2026-08-28

Every AI model price, context and end-of-life change detected on 2026-08-28.

ChangeModelDetail
END OF LIFEz-ai/glm-4.5vshuts down 2026-12-31 · 124 days left
END OF LIFE ANNOUNCEDmoonshotai/kimi-k2.5- -> 2026-08-31
END OF LIFE MOVED~z-ai/glm-latest2098-12-31 -> -
PRICE CHANGEdeepseek/deepseek-v3.2output +5.3% · cache read +3.5% · input +3.5%
PRICE CHANGEdeepseek/deepseek-v4-flashoutput +9.2% · cache read +9.2% · input +9.2%
PRICE CHANGEdeepseek/deepseek-v4-prooutput -14.8% · cache read -14.8% · input -14.8%
PRICE CHANGEdeepseek/deepseek-v4-pro-0813output -41.2% · cache read -41.2% · overrides now charged · input -41.2%
PRICE CHANGEgoogle/gemini-3.7-flashaudio +100.0% · output +100.0% · image +100.0% · input_audio_cache +100.0% · cache read +100.0% · cache write +100.0% · internal_reasoning +100.0% · input +100.0%
PRICE CHANGEmoonshotai/kimi-k2.5output +33.3% · cache read +42.9% · input +33.3%
PRICE CHANGEmoonshotai/kimi-k2.7-codecache read -5.3% · input -1.5%
PRICE CHANGEnvidia/nemotron-3-nano-30b-a3bcache read -16.7%
PRICE CHANGEnvidia/nemotron-3-ultra-550b-a55boutput -38.9% · cache read -50.0% · input -16.7%
PRICE CHANGEopenai/gpt-4.1-nanocache read +20.0%
PRICE CHANGEqwen/qwen3-235b-a22b-2507output -36.4% · cache read now charged · input -2.8%
PRICE CHANGEqwen/qwen3-vl-30b-a3b-instructoutput +15.4% · input +15.4%
PRICE CHANGEqwen/qwen3.5-122b-a10boutput +15.4% · input +11.5%
PRICE CHANGEqwen/qwen3.6-27boutput +12.5% · cache read now charged · input +87.5%
PRICE CHANGEqwen/qwen3.6-35b-a3boutput -10.0% · input -28.6%
PRICE CHANGEqwen/qwen3.8-2.4t-a95bcache read -20.0%
PRICE CHANGEz-ai/glm-4.6output -12.5% · cache read -20.0% · input -14.0%
PRICE CHANGE~deepseek/deepseek-v4-flash-latestoutput +33.3%
PRICE CHANGE~google/gemini-flash-latestaudio +100.0% · output +100.0% · image +100.0% · input_audio_cache +100.0% · cache read +100.0% · cache write +100.0% · internal_reasoning +100.0% · input +100.0%
PRICE CHANGE~z-ai/glm-latestinput -10.7%
NEW MODELinclusionai/ling-3.0-flash-fin:freecontext 262,144
NEW MODELmistralai/codestral-2508:batchcontext 256,000
NEW MODELmistralai/ministral-8b-2512:batchcontext 262,144
NEW MODELmistralai/mistral-large-2512:batchcontext 262,144
NEW MODELmistralai/mistral-medium-3-5:batchcontext 262,144
NEW MODELmistralai/mistral-medium-3.1:batchcontext 131,072
NEW MODELmistralai/mistral-small-2603:batchcontext 262,144
NEW MODELqwen/qwen3.8-2.4t-a95b:batchcontext 1,010,000
NEW MODELqwen/qwen3.8-flashcontext 1,000,000
NEW MODELtencent/hy4-previewcontext 1,048,576
MODEL REMOVEDmistralai/ministral-8b
MODEL REMOVEDopenai/gpt-5-codex:batch
MODEL REMOVEDthedrummer/rocinante-12b
FROM THE SOURCESam Altman @samai think we should do another party for our next model release, the 5.5 party was a lot of fun. what would make the next one awesome?
FROM THE SOURCEElon Musk @elonmuskShare your Grok @Bot with others! https://t.co/hY9uqE2tsb
FROM THE SOURCETibo @thsottiauxWhat is the most ambitious task you’ve given ChatGPT Work?
FROM THE SOURCETibo @thsottiaux@buildwithrajath Always improving. Team working hard on infra and usage optimizations
FROM THE SOURCETibo @thsottiaux@TokenGremlin Luna is used to give the session a title. A bit of a pass “you pass the butter” situation
FROM THE SOURCEAnthropic @AnthropicAIClaude can reliably fix measurable misalignment. But subtle or rare failures may have no benchmark at all—so everything hinges on measuring the right things. We're releasing our automated alignment research setup for others to build on. Ful
FROM THE SOURCEAnthropic @AnthropicAICould a model one day align its stronger successors? As a first test, we had Sonnet 5 post-train an early checkpoint of Opus 4.8, a more capable model. It reached safety scores approaching those of production Opus 4.8, which went through ou
FROM THE SOURCEAnthropic @AnthropicAIAcross 10 alignment failures, Claude reliably improved safety scores without degrading capabilities. Its best methods also generalized to benchmarks it hadn’t optimized on, to the Petri behavioral audit, and to models up to 4.7x larger. htt
FROM THE SOURCEAnthropic @AnthropicAIClaude “hill-climbed” safety benchmarks for common misalignments like deception or sycophancy, with one constraint: it had to preserve general capabilities. We then tested its best methods on held-out benchmarks to see if they'd generalize.
FROM THE SOURCEAnthropic @AnthropicAINew Fellows Research: Can Claude autonomously align other AIs? We gave Claude 48 hours and 1 GPU to improve the alignment of small models. It researched and proposed methods, then trained and tested the models on its own. It worked surprisi
OUTAGEElevated errors on Claude Code and Claude Cowork
ANNOUNCEMENTGemini Omni 1.1 Flash lets you build with more control
ANNOUNCEMENTPiloting the world's first double-blind AI evaluations
RELEASEv1.2.0
RELEASEv2.1.251
RELEASE0.150.1
RELEASElangchain-core==1.6.1
RELEASEv0.33.1
RELEASEv3.5.0
ANNOUNCEMENTEnterprise-managed settings now support autoUpdate for plugin marketplaces
ANNOUNCEMENTGlobal model policy generally available
ANNOUNCEMENTActions retention will cover checks, workflow runs, and statuses
ANNOUNCEMENTClose all open contributions authored by a blocked user
ANNOUNCEMENTCopilot code review: Resolution reasons and expanded capabilities
ANNOUNCEMENTGitHub Classroom deprecated
ANNOUNCEMENTBetter label management on issues is generally available
ANNOUNCEMENTUpcoming changes to GitHub Copilot policies and billing
ANNOUNCEMENTGitHub Copilot app for Beginners: Automate Dependabot pull request triage
ANNOUNCEMENTOpenClaw went viral. Meet the maintainers building and securing it.
OUTAGEDisruption with GitHub Billing
OUTAGEIncident with Actions and Pull Requests
OUTAGEIncident with Copilot AI Model Providers
WEIGHT_RELEASEgoogle/timesfm-3.0-pytorch
ANNOUNCEMENTExpanding OpenAI’s presence in Brazil
ANNOUNCEMENTThe Hugging Face incident and the road ahead
ANNOUNCEMENTIntroducing Intelligence Age
ANNOUNCEMENTSupporting Thailand’s next generation of AI startups
ANNOUNCEMENTBetter answers, broader thinking: What students gain from ChatGPT and critical-thinking training
OUTAGEUsers may experience an increase in error rates in Workspace Agents and ChatGPT Work on Web and Mobile
VARIANT DELISTEDmoonshotai/kimi-k2.7-code:batchthe ':batch' variant is no longer listed - moonshotai/kimi-k2.7-code itself is still there
VARIANT DELISTEDopenai/gpt-3.5-turbo:batchthe ':batch' variant is no longer listed - openai/gpt-3.5-turbo itself is still there
VARIANT DELISTEDopenai/gpt-4-turbo:batchthe ':batch' variant is no longer listed - openai/gpt-4-turbo itself is still there
VARIANT DELISTEDopenai/gpt-4.1-mini:batchthe ':batch' variant is no longer listed - openai/gpt-4.1-mini itself is still there
VARIANT DELISTEDopenai/gpt-4.1-nano:batchthe ':batch' variant is no longer listed - openai/gpt-4.1-nano itself is still there
VARIANT DELISTEDopenai/gpt-4.1:batchthe ':batch' variant is no longer listed - openai/gpt-4.1 itself is still there
VARIANT DELISTEDopenai/gpt-4o-mini:batchthe ':batch' variant is no longer listed - openai/gpt-4o-mini itself is still there
VARIANT DELISTEDopenai/gpt-4o:batchthe ':batch' variant is no longer listed - openai/gpt-4o itself is still there
VARIANT DELISTEDopenai/gpt-5-mini:batchthe ':batch' variant is no longer listed - openai/gpt-5-mini itself is still there
VARIANT DELISTEDopenai/gpt-5-nano:batchthe ':batch' variant is no longer listed - openai/gpt-5-nano itself is still there
VARIANT DELISTEDopenai/gpt-5-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.1:batchthe ':batch' variant is no longer listed - openai/gpt-5.1 itself is still there
VARIANT DELISTEDopenai/gpt-5.2-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5.2-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.2:batchthe ':batch' variant is no longer listed - openai/gpt-5.2 itself is still there
VARIANT DELISTEDopenai/gpt-5.4-mini:batchthe ':batch' variant is no longer listed - openai/gpt-5.4-mini itself is still there
VARIANT DELISTEDopenai/gpt-5.4-nano:batchthe ':batch' variant is no longer listed - openai/gpt-5.4-nano itself is still there
VARIANT DELISTEDopenai/gpt-5.4-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5.4-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.4:batchthe ':batch' variant is no longer listed - openai/gpt-5.4 itself is still there
VARIANT DELISTEDopenai/gpt-5.5-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5.5-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.5:batchthe ':batch' variant is no longer listed - openai/gpt-5.5 itself is still there
VARIANT DELISTEDopenai/gpt-5.6-luna-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5.6-luna-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.6-luna:batchthe ':batch' variant is no longer listed - openai/gpt-5.6-luna itself is still there
VARIANT DELISTEDopenai/gpt-5.6-sol-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5.6-sol-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.6-sol:batchthe ':batch' variant is no longer listed - openai/gpt-5.6-sol itself is still there
VARIANT DELISTEDopenai/gpt-5.6-terra-pro:batchthe ':batch' variant is no longer listed - openai/gpt-5.6-terra-pro itself is still there
VARIANT DELISTEDopenai/gpt-5.6-terra:batchthe ':batch' variant is no longer listed - openai/gpt-5.6-terra itself is still there
VARIANT DELISTEDopenai/gpt-5:batchthe ':batch' variant is no longer listed - openai/gpt-5 itself is still there
VARIANT DELISTEDopenai/o1-pro:batchthe ':batch' variant is no longer listed - openai/o1-pro itself is still there
VARIANT DELISTEDopenai/o1:batchthe ':batch' variant is no longer listed - openai/o1 itself is still there
VARIANT DELISTEDopenai/o3-mini-high:batchthe ':batch' variant is no longer listed - openai/o3-mini-high itself is still there
VARIANT DELISTEDopenai/o3-mini:batchthe ':batch' variant is no longer listed - openai/o3-mini itself is still there
VARIANT DELISTEDopenai/o3-pro:batchthe ':batch' variant is no longer listed - openai/o3-pro itself is still there
VARIANT DELISTEDopenai/o3:batchthe ':batch' variant is no longer listed - openai/o3 itself is still there
VARIANT DELISTEDopenai/o4-mini-high:batchthe ':batch' variant is no longer listed - openai/o4-mini-high itself is still there
VARIANT DELISTEDopenai/o4-mini:batchthe ':batch' variant is no longer listed - openai/o4-mini itself is still there
VARIANT DELISTEDz-ai/glm-5.2:batchthe ':batch' variant is no longer listed - z-ai/glm-5.2 itself is still there
CONTEXT CHANGEDgoogle/gemma-3-27b-it262144 -> 131072
CONTEXT CHANGEDkwaipilot/kat-coder-pro-v2.5256000 -> 262144
PARAMETERS CHANGEDkwaipilot/kat-coder-pro-v2.5added frequency_penalty, logit_bias, min_p, repetition_penalty, seed, top_k
CONTEXT CHANGEDmistralai/voxtral-small-24b-250732000 -> 32768
CONTEXT CHANGEDnvidia/nemotron-3-ultra-550b-a55b512288 -> 262144
PARAMETERS CHANGEDqwen/qwen3-next-80b-a3b-thinkingremoved logit_bias
CONTEXT CHANGEDz-ai/glm-5.31048576 -> 1310720
PARAMETERS CHANGEDz-ai/glm-5.3added frequency_penalty, logit_bias, logprobs, min_p, presence_penalty, repetition_penalty, seed, stop, structured_outputs, top_logprobs
CONTEXT CHANGED~z-ai/glm-latest1048576 -> 1310720
PARAMETERS CHANGED~z-ai/glm-latestadded frequency_penalty, logit_bias, logprobs, min_p, presence_penalty, repetition_penalty, seed, stop, structured_outputs, top_logprobs

Previous 2026-08-26 · Next 2026-08-29