# Models & prices news — updated daily | llmnet Apps

[← Apps overview](/en/)

# Models & prices news

By Ivo Donker · October 28, 2026 — created with AI assistance (writer: DeepSeek V4.1 Flash; critic: Gemini 3.8 Flash)

All the news about Models & prices from our daily newsletter: 136 items, newest first. Headlines come from the source; the Dutch interpretation is machine-written.

## Wednesday, October 28, 2026

- [Ran a quick test with @OpenGatewayai — one API for every LLM; straightforward pricing and solid docs. Exactly what I needed for model access](https://x.com/michaela271185/status/2115015219802210501) · X · 2026-10-27OpenGateway.ai offers one API for all LLMs with clear pricing and documentation, which gives indie‑developers an easy way to use models. · [newsletter](/en/nieuwsbrief-2026-10-28)

- [I made a dated changelog of LLM API price changes (free, no login)](https://www.reddit.com/r/SideProject/comments/1xgwtw6/i_made_a_dated_changelog_of_llm_api_price_changes/) · Reddit · 2026-10-27A free changelog of LLM‑API‑price changes is available without login, which helps indie‑developers with cost planning. · [newsletter](/en/nieuwsbrief-2026-10-28)

- [MemCurrent slashes token use 12X across 114 tasks. That makes persistent-memory agents far cheaper, pressuring LLM providers on per-token pr](https://x.com/murtuza_merc/status/2113670066147582166) · X · 2026-10-23MemCurrent reduces token usage by 12× across 114 tasks, making persistent‑memory agents cheaper for indie‑developers. · [newsletter](/en/nieuwsbrief-2026-10-28)

- [I built a site to compare LLM benchmarks, token pricing, and VRAM requirements in one place](https://www.reddit.com/r/AIToolsPerformance/comments/1xejtfu/i_built_a_site_to_compare_llm_benchmarks_token/) · Reddit · 2026-10-24A website compares LLM‑benchmarks, token prices and VRAM requirements, which helps indie‑developers choose a model. · [newsletter](/en/nieuwsbrief-2026-10-28)

## Tuesday, October 27, 2026

- [NIAH and RULER benchmarks are not good metrics for evaluating how models actually operate on long context. Looking back a bit in time, “long](https://x.com/ZoeyLi20/status/2114848109826293888) · X · 2026-10-26The article argues that the NIAH and RULER benchmarks are unreliable for long context, which is relevant for indie‑developers who need to estimate model performance in their own applications. · [newsletter](/en/nieuwsbrief-2026-10-27)

## Monday, October 26, 2026

- [Are traditional app subscription models breaking down in the AI era? How are you handling pricing as an indie dev?](https://www.reddit.com/r/AppBusiness/comments/1xei2h6/are_traditional_app_subscription_models_breaking/) · Reddit · 2026-10-24This item discusses changes in app subscriptions in the AI era, relevant for indie developers who want to price their software. · [newsletter](/en/nieuwsbrief-2026-10-26)

- [Same price per token, 2.7x the bill: what I learned trying to forecast my LLM API costs](https://www.reddit.com/r/LocalLLM/comments/1xcdmc5/same_price_per_token_27x_the_bill_what_i_learned/) · Reddit · 2026-10-22This item covers unexpectedly high LLM API costs despite equal token prices, important for the budget of indie developers. · [newsletter](/en/nieuwsbrief-2026-10-26)

- [I built a site to compare LLM benchmarks, token pricing, and VRAM requirements in one place](https://www.reddit.com/r/AIToolsPerformance/comments/1xejtfu/i_built_a_site_to_compare_llm_benchmarks_token/) · Reddit · 2026-10-24This item introduces a comparison site for LLM benchmarks and token prices, useful for indie developers who want to choose the right models. · [newsletter](/en/nieuwsbrief-2026-10-26)

- [I built a site to compare LLM benchmarks, token pricing, and VRAM requirements in one place](https://www.reddit.com/r/OpenAI/comments/1xejs8i/i_built_a_site_to_compare_llm_benchmarks_token/) · Reddit · 2026-10-24This item introduces a comparison site for LLM benchmarks and token prices, useful for indie developers who want to choose the right models. · [newsletter](/en/nieuwsbrief-2026-10-26)

## Friday, October 23, 2026

- [Same price per token, 2.7x the bill: what I learned trying to forecast my LLM API costs](https://www.reddit.com/r/LocalLLM/comments/1xcdmc5/same_price_per_token_27x_the_bill_what_i_learned/) · Reddit · 2026-10-22This is about predicting LLM API costs and token usage, which directly affects the operating costs of an indie app. · [newsletter](/en/nieuwsbrief-2026-10-23)

- [@grok so yeah. its a weaponized llm. ffs. okay. so i can get a subscription? From grok: ...... Yes, you can get a subscription. https://Abli](https://x.com/WLWeertman/status/2113159879192097271) · X · 2026-10-22This concerns information about a subscription to the LLM Grok, which may be relevant for choosing the right AI provider. · [newsletter](/en/nieuwsbrief-2026-10-23)

- [Send your LLM fewer tokens 🔎 Cohere launched Rerank 5 in Pro and Fast tiers: Pro ranked first on 39 of 46 benchmark datasets, and new Contex](https://x.com/galaxper/status/2113273615978287455) · X · 2026-10-22This is the launch of Cohere Rerank 5 to send fewer tokens to LLMs, which can help reduce API costs. · [newsletter](/en/nieuwsbrief-2026-10-23)

- [A good quote from an R&D meeting yesterday: "The frontier models are unstoppable." This wasn't a statement about security or a black-pill/wh](https://x.com/brandur/status/2112529201521717451) · X · 2026-10-20This is a quote about the progress of frontier models, which provides insight into overall technological progress. · [newsletter](/en/nieuwsbrief-2026-10-23)

## Thursday, October 22, 2026

- [LLM agents play Pokemon showdown](https://olivergrabner.com/research/pokemon/) · Hacker News · 2026-10-22LLM‑agents play Pokemon showdown, which shows indie‑developers how LLMs can be used in game environments. · [newsletter](/en/nieuwsbrief-2026-10-22)

- [Tired of messy spreadsheets for API pricing, so I built a tool to track and calculate LLM costs (modelanalysis.si)](https://www.reddit.com/r/OpenAI/comments/1xbk4ve/tired_of_messy_spreadsheets_for_api_pricing_so_i/) · Reddit · 2026-10-21A tool to track and calculate LLM‑costs replaces messy spreadsheets, which helps indie‑developers control their API‑budget. · [newsletter](/en/nieuwsbrief-2026-10-22)

- [Using LLM for confidence scoring is a bad idea](https://rumple.github.io/blog/2026/10/20/asking-llm-for-confidence.html) · Hacker News · 2026-10-21The post argues that using LLMs for confidence scoring is problematic, which warns indie‑developers about unreliable metrics. · [newsletter](/en/nieuwsbrief-2026-10-22)

- [feat(router): save the model router's logs at teardown](https://github.com/github/gh-aw-firewall/pull/10428) · GitHub · 2026-10-21The router‑feature saves model router logs on exit, which helps indie‑developers with debugging. · [newsletter](/en/nieuwsbrief-2026-10-22)

## Monday, October 19, 2026

- [Please help me understand: If one of the main points of competition is token prices ( lower = better ), then how would training LLM models be worthwhile in the long term?](https://www.reddit.com/r/aiwars/comments/1x7abel/please_help_me_understand_if_one_of_the_main/) · Reddit · 2026-10-16This is a question about the economic viability of LLM training with falling token prices, relevant for cost calculations of AI services. · [newsletter](/en/nieuwsbrief-2026-10-19)

- [[Available] Senior Full-Stack Engineer (6+ yrs) - React/Next.js, Go, Python, Laravel, AWS, AI/LLM - Remote (GMT+7, US/EU hours) - $35/hr or fixed-price](https://www.reddit.com/r/devjobs/comments/1x73uml/available_senior_fullstack_engineer_6_yrs/) · Reddit · 2026-10-16This is a profile of an available senior full-stack engineer, relevant for outsourcing or collaboration. · [newsletter](/en/nieuwsbrief-2026-10-19)

- [Show HN: Strot – Compile an LLM's intent once, run decisions locally](https://github.com/strot-ai/Project-STROT) · Hacker News · 2026-10-19This is a tool to compile an LLM's intent once and execute decisions locally, which saves on API costs. · [newsletter](/en/nieuwsbrief-2026-10-19)

## Sunday, October 18, 2026

- [Please help me understand: If one of the main points of competition is token prices ( lower = better ), then how would training LLM models be worthwhile in the long term?](https://www.reddit.com/r/aiwars/comments/1x7abel/please_help_me_understand_if_one_of_the_main/) · Reddit · 2026-10-16This raises questions about the long-term value of LLM training with falling token prices, which provides insight into the economics of AI services. · [newsletter](/en/nieuwsbrief-2026-10-18)

- [90% of US working hours are spent on tasks not covered well by AI benchmarks](https://www.mercor.com/research/labor-index/benchmark-map/) · Hacker News · 2026-10-16This reports that ninety percent of US working hours are spent on tasks that are poorly covered by AI benchmarks, which offers opportunities for specific niche software. · [newsletter](/en/nieuwsbrief-2026-10-18)

- [5/5 So recheck your model. When did you last check your model's output rate? Output price per 1M tokens for each model, collected daily: htt](https://x.com/Obolith42/status/2110415906106994701) · X · 2026-10-14This concerns a daily updated overview of output prices per million tokens, useful for monitoring the operating costs of AI integrations. · [newsletter](/en/nieuwsbrief-2026-10-18)

- [There's not a viable path to frontier lab profitability unless there's a viable path to pricing power: "The LLM Token Expenditure Index, a](https://x.com/trevornoren/status/2110756419364696262) · X · 2026-10-15This discusses the profitability and pricing power of frontier labs based on token spending, relevant for estimating future API costs. · [newsletter](/en/nieuwsbrief-2026-10-18)

## Saturday, October 17, 2026

- [90% of US working hours are spent on tasks not covered well by AI benchmarks](https://www.mercor.com/research/labor-index/benchmark-map/) · Hacker News · 2026-10-16Benchmarks currently miss ninety percent of US working hours, which highlights tasks where AI support remains limited. · [newsletter](/en/nieuwsbrief-2026-10-17)

- [Please help me understand: If one of the main points of competition is token prices ( lower = better ), then how would training LLM models be worthwhile in the long term?](https://www.reddit.com/r/aiwars/comments/1x7abel/please_help_me_understand_if_one_of_the_main/) · Reddit · 2026-10-16Declining token prices raise questions about long-term LLM training viability, which affects cost planning for tool integration. · [newsletter](/en/nieuwsbrief-2026-10-17)

- [Show HN: Parsimony – an open-source benchmark for how much code AI models add](https://parsimony.lasu.dev/) · Hacker News · 2026-10-13Parsimony is an open-source benchmark measuring AI code volume, which helps track how much code models add. · [newsletter](/en/nieuwsbrief-2026-10-17)

- [5/5 So recheck your model. When did you last check your model's output rate? Output price per 1M tokens for each model, collected daily: htt](https://x.com/Obolith42/status/2110415906106994701) · X · 2026-10-14Daily collected model output prices help indie developers track API expenses and select cost-effective models. · [newsletter](/en/nieuwsbrief-2026-10-17)

## Friday, October 16, 2026

- [5/5 So recheck your model. When did you last check your model's output rate? Output price per 1M tokens for each model, collected daily: htt](https://x.com/Obolith42/status/2110415906106994701) · X · 2026-10-14This concerns a daily updated overview of output prices per million tokens for AI models, which is relevant for controlling the API costs of a SaaS product. · [newsletter](/en/nieuwsbrief-2026-10-16)

- [Show HN: Parsimony – an open-source benchmark for how much code AI models add](https://parsimony.lasu.dev/) · Hacker News · 2026-10-13This concerns Parsimony, an open-source benchmark that measures how much code AI models add, which is relevant for measuring the efficiency of AI-generated code. · [newsletter](/en/nieuwsbrief-2026-10-16)

- [soooooo if i'm reading this right, google is shipping an LLM that will double in price in three months. Think this is the first time i've se](https://x.com/demian_ai/status/2110407786316108210) · X · 2026-10-14This concerns an announcement about a Google LLM whose price doubles within three months, which is relevant for budgeting AI costs. · [newsletter](/en/nieuwsbrief-2026-10-16)

- [[Available] Senior Full-Stack Engineer (6+ yrs) - React/Next.js, Go, Python, Laravel, AWS, AI/LLM - Remote (GMT+7, US/EU hours) - $35/hr or fixed-price](https://www.reddit.com/r/devjobs/comments/1x73uml/available_senior_fullstack_engineer_6_yrs/) · Reddit · 2026-10-16This concerns a profile of an available senior full-stack engineer, which may be relevant when outsourcing specialist development work. · [newsletter](/en/nieuwsbrief-2026-10-16)

## Tuesday, October 13, 2026

- [How would you price a Germany-hosted LLM API with support and reserved capacity?](https://www.reddit.com/r/SaaS/comments/1x3jh01/how_would_you_price_a_germanyhosted_llm_api_with/) · Reddit · 2026-10-11This discusses pricing for a Germany-hosted LLM API with support, useful when calculating costs for European API services. · [newsletter](/en/nieuwsbrief-2026-10-13)

- [Any tips for LLM API Pricing comparison? Building custom API app](https://www.reddit.com/r/CustomAI/comments/1x4iwv4/any_tips_for_llm_api_pricing_comparison_building/) · Reddit · 2026-10-13This concerns tips for comparing LLM API prices when building your own API app, relevant for cost control. · [newsletter](/en/nieuwsbrief-2026-10-13)

- [When you did LLM cost optimization, where did you draw the line on price vs quality?](https://www.reddit.com/r/Development/comments/1x1e6ys/when_you_did_llm_cost_optimization_where_did_you/) · Reddit · 2026-10-09This is about weighing price versus quality in LLM cost optimisation, useful for budgeting AI features. · [newsletter](/en/nieuwsbrief-2026-10-13)

- [After years of claiming, fruitlessly, that I would be behind the curve, the same crowd now consults me to clean up their own LLM mess](https://www.reddit.com/r/BetterOffline/comments/1x3l1fn/after_years_of_claiming_fruitlessly_that_i_would/) · Reddit · 2026-10-11This concerns the experience that people who were previously lagging behind are now asking for advice about LLM problems, which indicates changing expertise in the market. · [newsletter](/en/nieuwsbrief-2026-10-13)

## Monday, October 12, 2026

- [How would you price a Germany-hosted LLM API with support and reserved capacity?](https://www.reddit.com/r/SaaS/comments/1x3jh01/how_would_you_price_a_germanyhosted_llm_api_with/) · Reddit · 2026-10-11This item asks a question about pricing a Germany-hosted LLM API, which is relevant for monetising your own AI services. · [newsletter](/en/nieuwsbrief-2026-10-12)

- [When you did LLM cost optimization, where did you draw the line on price vs quality?](https://www.reddit.com/r/Development/comments/1x1e6ys/when_you_did_llm_cost_optimization_where_did_you/) · Reddit · 2026-10-09This item discusses trade-offs in LLM cost optimisation between price and quality, which helps with budgeting AI features. · [newsletter](/en/nieuwsbrief-2026-10-12)

- [After years of claiming, fruitlessly, that I would be behind the curve, the same crowd now consults me to clean up their own LLM mess](https://www.reddit.com/r/BetterOffline/comments/1x3l1fn/after_years_of_claiming_fruitlessly_that_i_would/) · Reddit · 2026-10-11This item describes experience with cleaning up other people's AI implementations, which provides insight into common technical mistakes. · [newsletter](/en/nieuwsbrief-2026-10-12)

- [A reproducible cost-per-task method for LLM pricing: sticker price per token ranked models differently in my own two benchmarks too](https://www.reddit.com/r/OpenAI/comments/1x0q5u5/a_reproducible_costpertask_method_for_llm_pricing/) · Reddit · 2026-10-08This item covers a method for calculating cost per task for LLMs, which helps accurately estimate operational costs. · [newsletter](/en/nieuwsbrief-2026-10-12)

## Sunday, October 11, 2026

- [A reproducible cost-per-task method for LLM pricing: sticker price per token ranked models differently in my own two benchmarks too](https://www.reddit.com/r/OpenAI/comments/1x0q5u5/a_reproducible_costpertask_method_for_llm_pricing/) · Reddit · 2026-10-08This is a method for calculating the cost per task for LLMs, relevant for optimising API spend. · [newsletter](/en/nieuwsbrief-2026-10-11)

- [When you did LLM cost optimization, where did you draw the line on price vs quality?](https://www.reddit.com/r/Development/comments/1x1e6ys/when_you_did_llm_cost_optimization_where_did_you/) · Reddit · 2026-10-09This is a discussion about LLM cost optimisation, useful for balancing price and quality in applications. · [newsletter](/en/nieuwsbrief-2026-10-11)

- [Ask HN: Are you a subscribed LLM or a locally deployed open-source model?](https://news.ycombinator.com/item?id=50016644) · Hacker News · 2026-10-09This is a discussion about using subscription LLMs versus local open-source models for development. · [newsletter](/en/nieuwsbrief-2026-10-11)

- [Reading many rebuttals to this post suggests there is little understanding of how Chinese open-source models are able to reach the same leve](https://x.com/JaguarAnalytics/status/2109037443488039411) · X · 2026-10-10This is an analysis of how Chinese open-source models work, relevant for understanding global AI developments. · [newsletter](/en/nieuwsbrief-2026-10-11)

## Wednesday, October 7, 2026

- [Pricing and viability of indie apps post-AI](https://www.reddit.com/r/macapps/comments/1wvyd4s/pricing_and_viability_of_indie_apps_postai/) · Hacker News · 2026-10-05Discussion about pricing and viability of indie apps after AI; relevant for indie developers who need to adapt their business model. · [newsletter](/en/nieuwsbrief-2026-10-07)

- [An Old GPU Mining Rig With 9 GPUs Sporting 54GB VRAM Is An LLM User’s Dream; Buyer Purchased Hardware At A Throwaway Figure By Negotiating The Price Down To $35](https://www.reddit.com/r/NewMaxx/comments/1wzfp5j/an_old_gpu_mining_rig_with_9_gpus_sporting_54gb/) · Reddit · 2026-10-06An old GPU-mining rig with 9 GPUs and 54 GB VRAM is bought for $35; relevant for indie developers looking for cheap hardware for LLM experiments. · [newsletter](/en/nieuwsbrief-2026-10-07)

- [Ask HN: How come everyone is an LLM expert?](https://news.ycombinator.com/item?id=49980529) · Hacker News · 2026-10-06Question why everyone is an LLM expert; relevant for indie developers who need to assess their level of knowledge and market expectations. · [newsletter](/en/nieuwsbrief-2026-10-07)

- [Translate Copilot requests across model-supported wire APIs](https://github.com/github/gh-aw-firewall/pull/9523) · GitHub · 2026-10-07GitHub project translates Copilot requests about model-supported wire APIs; relevant for indie developers who want to integrate different AI models into their workflow. · [newsletter](/en/nieuwsbrief-2026-10-07)

## Tuesday, October 6, 2026

- [Jamba had a 2-week run. Cloudflare's Clef is 4x faster, cuts error rates in half, adds vision, and doubles context window. The pace of model](https://x.com/abeauvois/status/2107006886528577863) · X · 2026-10-05Cloudflare's Clef offers a four times faster LLM with half the error rate, vision support and double the context, which helps indie developers build more efficient applications · [newsletter](/en/nieuwsbrief-2026-10-06)

## Monday, October 5, 2026

- [LLM updates are starting to feel like iPhone releases. New model: slightly faster slightly smarter new pricing everyone argues for 3 days](https://x.com/dev_maims/status/2104941881444503714) · X · 2026-09-29The post is about a new LLM model that is slightly faster and smarter and has a new pricing model, which is relevant for indie developers because it can affect their costs and performance. · [newsletter](/en/nieuwsbrief-2026-10-05)

## Sunday, October 4, 2026

- [Instinct has raised $1 billion in Series C funding at a $10 billion valuation, with participation from Sequoia Capital, Benchmark and Coatue](https://x.com/bipsiq/status/2105738635182227877) · X · 2026-10-01This concerns a funding round of Instinct of 1亿美元, which provides insight into market investments. · [newsletter](/en/nieuwsbrief-2026-10-04)

- [Most complex multi-agent frameworks are pure overhead Recent ablation studies on ML engineering benchmarks show that elaborate orchestrators](https://x.com/LouieMota88/status/2105991448260956604) · X · 2026-10-02This shows that complex multi-agent frameworks cause unnecessary overhead, which helps with streamlining architectures. · [newsletter](/en/nieuwsbrief-2026-10-04)

- [Hey @X algorithm, find my people 🎯 Let's connect 🤝 🎯 Goal: first 150 followers. I am currently at 103 followers. Help me close the gap. Week](https://x.com/som_dutt_/status/2106439543373373536) · X · 2026-10-03This is a call to reach followers on X, which is relevant for social network growth. · [newsletter](/en/nieuwsbrief-2026-10-04)

- [after 3 yrs on framer, I just cancelled it today and launched the new https://www.baked.design/ using claude, github & cloudflare > https://](https://x.com/nickbakeddesign/status/2105675634727141538) · X · 2026-10-01This concerns a website launched with Claude, GitHub and Cloudflare after cancelling Framer, useful for alternative workflows. · [newsletter](/en/nieuwsbrief-2026-10-04)

## Saturday, October 3, 2026

- [Benchmarking AI Agent Performance: Metrics and Frameworks https://msinformationtech.blogspot.com/2026/10/benchmarking-ai-agent-performance.h](https://x.com/MSInfoTech5/status/2105660764652478916) · X · 2026-10-01This article discusses metrics and frameworks for benchmarking AI agents, which is relevant for measuring agent performance. · [newsletter](/en/nieuwsbrief-2026-10-03)

- [Uncensored and Offensive Security AI Models Benchmark](https://github.com/JoasASantos/Offensive-Security-AI-Models) · Hacker News · 2026-09-29This benchmark focuses on uncensored and offensive security AI models, which is relevant for security researchers. · [newsletter](/en/nieuwsbrief-2026-10-03)

- [Your AI agent's model changes more than its answers. It also changes the speed, the reasoning, and the AI Credits each task can consume. In](https://x.com/HatcherLabs/status/2105911513445789716) · X · 2026-10-02This article covers how model changes affect speed, reasoning ability and costs, which is relevant for API cost management. · [newsletter](/en/nieuwsbrief-2026-10-03)

- [So for $PODM @pod_miner I *believe* it Runs LLM inference on the https://UsePod.ai marketplace Finds the cheapest GPU boxes on https://Vast.](https://x.com/ahhhcryptoblows/status/2106042194230436077) · X · 2026-10-02This system performs LLM inference and looks for cheap GPUs, which is relevant for reducing the costs of AI hosting. · [newsletter](/en/nieuwsbrief-2026-10-03)

## Thursday, October 1, 2026

- [Fix Copilot session state collection](https://github.com/github/gh-aw/pull/64461) · GitHub · 2026-09-30This is a bug fix for collecting Copilot session status. · [newsletter](/en/nieuwsbrief-2026-10-01)

- ["As a Language Model": Chat Template Switches LLM Self-Referential Voice](https://arxiv.org/abs/2609.25021) · Hacker News · 2026-09-27This is about chat templates that change the self-referential voice of LLMs. · [newsletter](/en/nieuwsbrief-2026-10-01)

- [MicroLLM Lab – Try 7 tiny LLM's in the browser](https://stateofutopia.com/experiments/microllmlab/) · Hacker News · 2026-09-28This is a lab environment for trying seven small LLMs in the browser. · [newsletter](/en/nieuwsbrief-2026-10-01)

- [Model routing: make the selection advisory instead of pinning requests](https://github.com/github/gh-aw-firewall/pull/9272) · GitHub · 2026-09-30This concerns model routing where selection becomes advisory instead of fixed. · [newsletter](/en/nieuwsbrief-2026-10-01)

## Tuesday, September 29, 2026

- [My current essentials for macOS](https://www.reddit.com/r/macapps/comments/1wnc8gi/my_current_essentials_for_macos/) · Reddit · 2026-09-22A Reddit post with an overview of essential tools for macOS, useful for indie developers who work on this platform. · [newsletter](/en/nieuwsbrief-2026-09-29)

- ["As a Language Model": Chat Template Switches LLM Self-Referential Voice](https://arxiv.org/abs/2609.25021) · Hacker News · 2026-09-27An article about a chat template that changes the self-referential voice of an LLM, relevant for indie developers who want to adjust the output style of their models. · [newsletter](/en/nieuwsbrief-2026-09-29)

- [The LLM Job Paradox](https://blog.nilesh.io/post/llms-and-jobs) · Hacker News · 2026-09-27A discussion about the LLM Job Paradox, which gives indie developers insight into the impact of LLMs on employment. · [newsletter](/en/nieuwsbrief-2026-09-29)

- [Share compose-generator test fixture setup via a test helper](https://github.com/github/gh-aw-firewall/pull/9159) · GitHub · 2026-09-28A GitHub contribution that shares a test helper for compose generator fixtures, allowing indie developers to reuse test setup. · [newsletter](/en/nieuwsbrief-2026-09-29)

## Sunday, September 27, 2026

- [💰 LLM pricing isn’t just about cost per request. It’s about cost per token × input/output tokens × request volume. A model that looks cheape](https://x.com/Badrinath248/status/2103908635088925051) · X · 2026-09-26 · [newsletter](/en/nieuwsbrief-2026-09-27)

- [Tuesday morning AI FACTS roundup 🤖 22.09 - 🚀 xAI unveils the Grok 4.7 model with longer agentic operation and pricing of $2 / $6 per 1M toke](https://x.com/AIDailyNews_2x/status/2102300955320389884) · X · 2026-09-22 · [newsletter](/en/nieuwsbrief-2026-09-27)

- [If you use a full max plan at API pricing that’s ~$8k/mo or $96k/year. Thats why most big companies who can’t get max plans are slashing LLM](https://x.com/barbinbrad/status/2103817314865164700) · X · 2026-09-26 · [newsletter](/en/nieuwsbrief-2026-09-27)

- [refactor(llm): replace litellm model-cost table with vendored catalog](https://github.com/onyx-dot-app/onyx/pull/15003) · GitHub · 2026-09-22 · [newsletter](/en/nieuwsbrief-2026-09-27)

## Saturday, September 26, 2026

- [Tuesday morning AI FACTS roundup 🤖 22.09 - 🚀 xAI unveils the Grok 4.7 model with longer agentic operation and pricing of $2 / $6 per 1M toke](https://x.com/AIDailyNews_2x/status/2102300955320389884) · X · 2026-09-22xAI introduces the Grok 4.7 model with longer agent operation and a price of $2/$6 per million tokens, relevant for indie developers looking for affordable LLM capacity. · [newsletter](/en/nieuwsbrief-2026-09-26)

## Friday, September 25, 2026

- [refactor(llm): replace litellm model-cost table with vendored catalog](https://github.com/onyx-dot-app/onyx/pull/15003) · GitHub · 2026-09-22 · [newsletter](/en/nieuwsbrief-2026-09-25)

- [Tuesday morning AI FACTS roundup 🤖 22.09 - 🚀 xAI unveils the Grok 4.7 model with longer agentic operation and pricing of $2 / $6 per 1M toke](https://x.com/AIDailyNews_2x/status/2102300955320389884) · X · 2026-09-22 · [newsletter](/en/nieuwsbrief-2026-09-25)

- [Global LLM price war continues - OpenAI has introduced two new tiers—GPT‑6 Sol and GPT‑6 Luna—to its GPT‑6 series, with API pricing reduced](https://x.com/HedgeyeTech/status/2102584091019215312) · X · 2026-09-23 · [newsletter](/en/nieuwsbrief-2026-09-25)

- [LiteLLM replacement G3: gateway streaming + usage/spend recording (UNKNOWN never 0)](https://github.com/jaylfc/taOS/pull/3182) · GitHub · 2026-09-24 · [newsletter](/en/nieuwsbrief-2026-09-25)

## Tuesday, September 22, 2026

- [docs: add a model loading and startup acceleration guide](https://github.com/llm-d/llm-d/pull/2540) · GitHub · 2026-09-21docs: add a model loading and startup acceleration guide adds a guide for faster model loading, which can save indie developers time at startup. · [newsletter](/en/nieuwsbrief-2026-09-22)

- [[openrouter] Missing model: google/gemini-3.8-flash](https://github.com/langgenius/dify-official-plugins/issues/3894) · GitHub · 2026-09-17[openrouter] Missing model: google/gemini-3.8-flash reports that the model google/gemini-3.8-flash is unavailable, which should let indie developers know that they need to look for an alternative. · [newsletter](/en/nieuwsbrief-2026-09-22)

- [How to Write with an LLM](https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/) · Hacker News · 2026-09-17How to Write with an LLM offers guidelines for writing with a large language model, which can support indie developers with content generation. · [newsletter](/en/nieuwsbrief-2026-09-22)

- [[FEATURE] configure model deployments with non standard names to use the standard name for cost and capabilities discovery](https://github.com/crmne/ruby_llm/issues/940) · GitHub · 2026-09-17[FEATURE] configure model deployments with non standard names to use the standard name for cost and capabilities discovery describes a feature to link non-standard model names to standard names for cost and capability detection · [newsletter](/en/nieuwsbrief-2026-09-22)

## Sunday, September 20, 2026

- [Deploy an LLM as an agent instead of a prompt, and watch your token count jump as much as 10-100x. Nothing about the model changes. Nothing](https://x.com/QumulusAI/status/2100571022751674449) · X · 2026-09-17The article warns that deploying an LLM as an agent instead of a prompt can greatly increase token usage, which is important for indie developers with a limited budget. · [newsletter](/en/nieuwsbrief-2026-09-20)

- [[openrouter] Missing model: google/gemini-3.8-flash](https://github.com/langgenius/dify-official-plugins/issues/3894) · GitHub · 2026-09-17The notice about the missing model google/gemini‑3.8‑flash lets indie developers know that they need to find an alternative or handle the error. · [newsletter](/en/nieuwsbrief-2026-09-20)

- [How to Write with an LLM](https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/) · Hacker News · 2026-09-17The guide explains how to write with an LLM, which can support indie developers in generating text for their projects. · [newsletter](/en/nieuwsbrief-2026-09-20)

- [The turbo-fast jump in LLM quality we’ve seen over the past year doesn’t actually move the AI industry the way most people think. It’s not l](https://x.com/kamilkwapiszpl/status/2100851820922814782) · X · 2026-09-18The article states that the rapid quality increase of LLMs does not change the industry in the expected way, which helps indie developers set realistic goals. · [newsletter](/en/nieuwsbrief-2026-09-20)

## Saturday, September 19, 2026

- [The turbo-fast jump in LLM quality we’ve seen over the past year doesn’t actually move the AI industry the way most people think. It’s not l](https://x.com/kamilkwapiszpl/status/2100851820922814782) · X · 2026-09-18The rapid improvement in LLM quality changes the AI market less than expected, which helps indie developers set realistic expectations. · [newsletter](/en/nieuwsbrief-2026-09-19)

- [feat(llm-catalog): add InclusionAI Ling 3.0 Flash VL](https://github.com/Significant-Gravitas/AutoGPT/pull/14683) · GitHub · 2026-09-17InclusionAI Ling 3.0 Flash VL has been added to llm‑catalog, allowing indie developers to use a new model. · [newsletter](/en/nieuwsbrief-2026-09-19)

- [How to Write with an LLM](https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/) · Hacker News · 2026-09-17Article about how to write with an LLM offers indie developers practical tips for content generation. · [newsletter](/en/nieuwsbrief-2026-09-19)

- [input costs, output is free - the opposite of typical llm pricing.](https://x.com/adenshepard/status/2100756661585461255) · X · 2026-09-18This LLM charges only for input and offers free output, which can reduce costs for indie developers. · [newsletter](/en/nieuwsbrief-2026-09-19)

## Friday, September 18, 2026

- [feat(backend): add Unbiased Pareto to LLM catalog](https://github.com/Significant-Gravitas/AutoGPT/pull/14697) · GitHub · 2026-09-18 · [newsletter](/en/nieuwsbrief-2026-09-18)

- [Deploy an LLM as an agent instead of a prompt, and watch your token count jump as much as 10-100x. Nothing about the model changes. Nothing](https://x.com/QumulusAI/status/2100571022751674449) · X · 2026-09-17 · [newsletter](/en/nieuwsbrief-2026-09-18)

- [feat(llm-catalog): add InclusionAI Ling 3.0 Flash VL](https://github.com/Significant-Gravitas/AutoGPT/pull/14683) · GitHub · 2026-09-17 · [newsletter](/en/nieuwsbrief-2026-09-18)

- [How to Write with an LLM](https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/) · Hacker News · 2026-09-17 · [newsletter](/en/nieuwsbrief-2026-09-18)

## Wednesday, September 16, 2026

- [Building your first LLM API call in Python (step by step)](https://heymeraki.substack.com/p/aie_10-building-it) · Hacker News · 2026-09-15 · [newsletter](/en/nieuwsbrief-2026-09-16)

- [ok everyone, i took one for the team here's the data we all wanted to see - real token value of each LLM subscription, empirically measured](https://x.com/kunchenguid/status/2098256018836963382) · X · 2026-09-11 · [newsletter](/en/nieuwsbrief-2026-09-16)

- [Show HN: Determinstic LLM inference for lowest price Gemma 4, with Windows XP](https://www.tokendelivery.ai/) · Hacker News · 2026-09-12 · [newsletter](/en/nieuwsbrief-2026-09-16)

- [Admin: model configuration per stage, and what each stage has spent — needs a design session](https://github.com/tadasant/motet/issues/92) · GitHub · 2026-09-13 · [newsletter](/en/nieuwsbrief-2026-09-16)

## Tuesday, September 15, 2026

- [ok everyone, i took one for the team here's the data we all wanted to see - real token value of each LLM subscription, empirically measured](https://x.com/kunchenguid/status/2098256018836963382) · X · 2026-09-11 · [newsletter](/en/nieuwsbrief-2026-09-15)

- [Introducing Colossus and the launch of $COLOSSUS 🗽 Colossus brings price discovery to AI inference through a marketplace for discounted LLM](https://x.com/colossuseco/status/2097327501047693531) · X · 2026-09-08 · [newsletter](/en/nieuwsbrief-2026-09-15)

- [Show HN: Determinstic LLM inference for lowest price Gemma 4, with Windows XP](https://www.tokendelivery.ai/) · Hacker News · 2026-09-12 · [newsletter](/en/nieuwsbrief-2026-09-15)

- [Admin: model configuration per stage, and what each stage has spent — needs a design session](https://github.com/tadasant/motet/issues/92) · GitHub · 2026-09-13 · [newsletter](/en/nieuwsbrief-2026-09-15)

## Sunday, September 13, 2026

- [Real-SWE: Benchmarking AI models on private, real-world, enterprise codebases](https://withspecific.com/benchmarks/real-swe) · Hacker News · 2026-09-12A benchmark (Real‑SWE) of AI models on private, enterprise codebases, relevant for indie developers who want to compare model performance. · [newsletter](/en/nieuwsbrief-2026-09-13)

- [Introducing Colossus and the launch of $COLOSSUS 🗽 Colossus brings price discovery to AI inference through a marketplace for discounted LLM](https://x.com/colossuseco/status/2097327501047693531) · X · 2026-09-08Introduction of Colossus, a marketplace for cheap LLM inference, relevant for indie developers who want to save costs on AI usage. · [newsletter](/en/nieuwsbrief-2026-09-13)

- [Another angle of my LLM coding leaderboard: price/quality. Two charts: 1. Cheaper LLMs up to $0.50 per prompt (API pricing) 2. More expensi](https://x.com/PovilasKorop/status/2098019567452787138) · X · 2026-09-10An analysis of price-quality of LLMs, with charts for cheap and more expensive models, useful for indie developers when choosing a model. · [newsletter](/en/nieuwsbrief-2026-09-13)

- [Admin: model configuration per stage, and what each stage has spent — needs a design session](https://github.com/tadasant/motet/issues/92) · GitHub · 2026-09-13A GitHub issue about model configuration per phase and cost overview, relevant for indie developers who want to manage their AI pipeline. · [newsletter](/en/nieuwsbrief-2026-09-13)

## Saturday, September 12, 2026

- [🚀 OpenBMB releases MiniCPM5-2B! A 2.5B parameter dense LLM setting a new 2B-class SOTA (53.9 avg). Features: • 131k context window • Outper](https://x.com/suyang339666/status/2097136536282415257) · X · 2026-09-08OpenBMB has released MiniCPM5-2B, a 2.5 B-parameter LLM with a context window of 131 k, which makes long-text applications possible for indie developers. · [newsletter](/en/nieuwsbrief-2026-09-12)

## Wednesday, September 9, 2026

- [Introducing Colossus and the launch of $COLOSSUS 🗽 Colossus brings price discovery to AI inference through a marketplace for discounted LLM](https://x.com/colossuseco/status/2097327501047693531) · X · 2026-09-08 · [newsletter](/en/nieuwsbrief-2026-09-09)

- [LLM inference pricing changes daily across platforms. Price Dashboard tracks 100 plus providers every 6 hours: - History for DeepSeek, K3 an](https://x.com/so_sthbryan/status/2097186668222759331) · X · 2026-09-08 · [newsletter](/en/nieuwsbrief-2026-09-09)

- [→ https://StandardCOmpute.com is a flat-rate, OpenAI-compatible LLM API for coding agents designed to lower AI costs via smart model routing](https://x.com/devhunt_/status/2097339569251217746) · X · 2026-09-08 · [newsletter](/en/nieuwsbrief-2026-09-09)

- [→ @FluxionAi_HQ is a unified AI API gateway that provides pay-as-you-go access to global and Chinese LLM models with discounted token pricin](https://x.com/devhunt_/status/2097339548111888701) · X · 2026-09-08 · [newsletter](/en/nieuwsbrief-2026-09-09)

## Sunday, September 6, 2026

- [We benchmarked Claude Fable 5.1 and Gemini 3.8 Flash on surgical intelligence: - The performance of the two new releases is spiky (e.g. goo](https://x.com/kskblv/status/2095246086218686610) · X · 2026-09-02It is a benchmark of Claude Fable 5.1 and Gemini 3.8 Flash on surgical intelligence, reporting inconsistent performance, which is relevant for indie developers who want to use these models for critical applications. · [newsletter](/en/nieuwsbrief-2026-09-06)

## Friday, September 4, 2026

- [We benchmarked Claude Fable 5.1 and Gemini 3.8 Flash on surgical intelligence: - The performance of the two new releases is spiky (e.g. goo](https://x.com/kskblv/status/2095246086218686610) · X · 2026-09-02The benchmark compares Claude Fable 5.1 and Gemini 3.8 Flash on surgical intelligence and reports inconsistent performance, which indie developers should consider when choosing a model. · [newsletter](/en/nieuwsbrief-2026-09-04)

## Thursday, September 3, 2026

- [We benchmarked Claude Fable 5.1 and Gemini 3.8 Flash on surgical intelligence: - The performance of the two new releases is spiky (e.g. goo](https://x.com/kskblv/status/2095246086218686610) · X · 2026-09-02An X post with benchmark results from Claude Fable 5.1 and Gemini 3.8 Flash on surgical intelligence, relevant for indie developers comparing model performance. · [newsletter](/en/nieuwsbrief-2026-09-03)

## Monday, August 31, 2026

- [Confession: with my LLM coding benchmarks, I'm pretty close to scenario when Terminal is ALWAYS running some evaluation of some model. Almo](https://x.com/PovilasKorop/status/2093686109205078350) · X · 2026-08-29The author indicates that their terminal continuously runs LLM evaluations, which gives indie developers insight into runtime performance. · [newsletter](/en/nieuwsbrief-2026-08-31)

## Sunday, August 30, 2026

- [Confession: with my LLM coding benchmarks, I'm pretty close to scenario when Terminal is ALWAYS running some evaluation of some model. Almos](https://x.com/PovilasKorop/status/2093686109205078350) · X · 2026-08-29The tweet indicates that the terminal continuously runs LLM evaluations, which makes indie developers aware of the possible compute load. · [newsletter](/en/nieuwsbrief-2026-08-30)

## Friday, August 28, 2026

- [-> AI labs spend zillions of dollars on data and compute -> Models get a little better -> "We beat the benchmark" -> Repeat over and over ag](https://x.com/Shaughnessy119/status/2090598587847577746) · X · 2026-08-21Short analysis of the cycle in which AI labs spend a lot of money, improve models and achieve benchmarks, relevant for indie developers considering AI costs. · [newsletter](/en/nieuwsbrief-2026-08-28)

## Wednesday, August 26, 2026

- [feat(aio): use Vercel AI Gateway reported cost for LLM generations](https://github.com/PostHog/posthog/pull/88457) · GitHub · 2026-08-25The aio‑feature uses the costs reported by Vercel AI Gateway for LLM generations, which helps indie developers monitor their spending. · [newsletter](/en/nieuwsbrief-2026-08-26)

- [If I had 6 months to master Context Engineering. I'd do this. Prompt engineering is dead. Context engineering is the new meta. Stage 1: LLM](https://x.com/suraj_sharma14/status/2090770893786009790) · X · 2026-08-21The author states that prompt engineering is outdated and that context engineering is the new focus, which can lead indie developers to better LLM integrations. · [newsletter](/en/nieuwsbrief-2026-08-26)

- [The real value of @OpenRouter is aggregating latency, uptime, throughput, pricing info across the 100+ hosted oss llm providers But you do n](https://x.com/rjchint/status/2092382541827436936) · X · 2026-08-25OpenRouter offers aggregation of latency, uptime, throughput and pricing information from hosted open-source LLM providers — according to the X post from more than 100 providers, although public sources mention around 400+ models from 60+ providers — which is useful for indie developers who want to compare providers. · [newsletter](/en/nieuwsbrief-2026-08-26)

- [consumer llm pricing](https://x.com/tilde_posting/status/2091746514578370648) · X · 2026-08-24There is information about consumer LLM prices, which helps indie developers with cost planning. · [newsletter](/en/nieuwsbrief-2026-08-26)

## Tuesday, August 25, 2026

- [Everyone is trying to searching how to save on LLM API costs, but people rarely understand how those API costs are setup in the first place.](https://x.com/Harshitp147/status/2091594136545550735) · X · 2026-08-23This points to a lack of understanding about how LLM API costs are structured, which is relevant for indie developers who want to control AI costs. · [newsletter](/en/nieuwsbrief-2026-08-25)

- [OCR It – pull text out of un-copyable documents for your LLM](https://github.com/thiagotigaz/ocr-it) · Hacker News · 2026-08-24This is an OCR tool to extract text from non-copyable documents for an LLM, which helps indie developers feed document data to AI. · [newsletter](/en/nieuwsbrief-2026-08-25)

- [consumer llm pricing](https://x.com/tilde_posting/status/2091746514578370648) · X · 2026-08-24This concerns consumer prices of LLMs, which helps indie developers estimate the costs for end users of AI functionality. · [newsletter](/en/nieuwsbrief-2026-08-25)

- [DeepSeek V4 Pro is GA: adaptive reasoning, Responses API, peak/off-peak pricing. Mixed benchmarks, but strong in cybersecurity. So what? A s](https://x.com/AxelWinterBkk/status/2090298297936486666) · X · 2026-08-20This announces DeepSeek V4 Pro with adaptive reasoning, a Responses API, peak and off-peak pricing and mixed benchmarks, which gives indie developers an additional LLM option. · [newsletter](/en/nieuwsbrief-2026-08-25)

## Monday, August 24, 2026

- [-> AI labs spend zillions of dollars on data and compute -> Models get a little better -> "We beat the benchmark" -> Repeat over and over ag](https://x.com/Shaughnessy119/status/2090598587847577746) · X · 2026-08-21The description of the cycle in which AI labs spend a lot of money on data and compute shows indie developers the context of model improvements. · [newsletter](/en/nieuwsbrief-2026-08-24)

## Sunday, August 23, 2026

- [I built a real-time LLM API pricing comparator — because I was tired of not knowing the actual cost difference between models](https://www.reddit.com/r/SideProject/comments/1vuom3w/i_built_a_realtime_llm_api_pricing_comparator/) · Reddit · 2026-08-21 · [newsletter](/en/nieuwsbrief-2026-08-23)

- [OpenAI dropped GPT-5.6 Sol pricing on August 20. Input is now $4 per million tokens, output $20 per million, 20% lower input and 33% lower o](https://x.com/Chad_Justice_AI/status/2091087034534306227) · X · 2026-08-22 · [newsletter](/en/nieuwsbrief-2026-08-23)

- [Alibaba releases Qwen3.8-27B open weights — a 27B multimodal model that beats Qwen3.7-Plus on several coding and agent benchmarks](https://www.reddit.com/r/AIGuild/comments/1vqduy0/alibaba_releases_qwen3827b_open_weights_a_27b/) · Reddit · 2026-08-17 · [newsletter](/en/nieuwsbrief-2026-08-23)

## Saturday, August 22, 2026

- [OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20%](https://www.reuters.com/technology/openai-cuts-developer-pricing-frontier-gpt-56-sol-model-by-more-than-20-2026-08-21/) · Hacker News · 2026-08-22 · [newsletter](/en/nieuwsbrief-2026-08-22)

- [Alibaba releases Qwen3.8-27B open weights — a 27B multimodal model that beats Qwen3.7-Plus on several coding and agent benchmarks](https://www.reddit.com/r/AIGuild/comments/1vqduy0/alibaba_releases_qwen3827b_open_weights_a_27b/) · Reddit · 2026-08-17 · [newsletter](/en/nieuwsbrief-2026-08-22)

- [What locked-in price and usage limits would make you switch LLM providers?](https://www.reddit.com/r/openrouter/comments/1vs0bqq/what_lockedin_price_and_usage_limits_would_make/) · Reddit · 2026-08-18 · [newsletter](/en/nieuwsbrief-2026-08-22)

## Friday, August 21, 2026

- [We burned 11.7B tokens to find the best cyber AI model](https://www.aikido.dev/blog/ai-model-benchmarks-aug-21-2026) · Hacker News · 2026-08-21 · [newsletter](/en/nieuwsbrief-2026-08-21)

- [feat(cognition): give Cognition its own provider identity](https://github.com/BerriAI/litellm/pull/37743) · GitHub · 2026-08-21 · [newsletter](/en/nieuwsbrief-2026-08-21)

- [feat(travelclick): add travelclick](https://github.com/mvanhorn/printing-press-library/pull/1775) · GitHub · 2026-08-21 · [newsletter](/en/nieuwsbrief-2026-08-21)

- [WhatsApp raised some API rates 300% and will end free messages soon](https://wha.tools/blog/whatsapp-business-pricing-investigation-2026) · Hacker News · 2026-08-21 · [newsletter](/en/nieuwsbrief-2026-08-21)

## Thursday, August 20, 2026

- [Most AI cost overruns are not a pricing problem. They are a routing problem. A classification task that could run on a small model runs on a](https://x.com/SuperwiseAI/status/2090169152258347257) · X · 2026-08-19 · [newsletter](/en/nieuwsbrief-2026-08-20)

- [New benchmark just dropped. Not another "can it solve the end of the world" exam. Not another biology or philosophy olympiad. This one answe](https://x.com/PhyByte/status/2090065283146760337) · X · 2026-08-19 · [newsletter](/en/nieuwsbrief-2026-08-20)

- [feat(llm): add moonshot/kimi-k3 to model prices and context window map](https://github.com/BerriAI/litellm/pull/37552) · GitHub · 2026-08-19 · [newsletter](/en/nieuwsbrief-2026-08-20)