AI model deprecations and retirement tracker

Context

200K

Max output

64K

Input / output

$3 / $15

Caveat: Deprecated September 30, 2026. The November 30, 2026 retirement applies to Anthropic-operated platforms (Claude API, Claude Platform on AWS, Microsoft Foundry); Amazon Bedrock and Google Cloud set their own dates.

Context

400K

Max output

128K

Input / output

$1.25 / $10

Pricing note: Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Maximum input is 272,000 tokens of the 400,000-token context window. OpenAI's model page now recommends GPT-6 Astra.

Context

400K

Max output

128K

Input / output

$0.25 / $2

Pricing note: Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Maximum input is 272,000 tokens of the 400,000-token context window.

Context

400K

Max output

128K

Input / output

$0.050 / $0.40

Pricing note: Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Maximum input is 272,000 tokens of the 400,000-token context window.

Context

400K

Max output

272K

Input / output

$15 / $120

Pricing note: No cached-input discount. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Responses API only; it supports only reasoning.effort high, and some requests may take several minutes.

Context

200K

Max output

100K

Input / output

$2 / $8

Pricing note: OpenAI's Batch table lists no cached-input rate for o3. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Context

200K

Max output

100K

Input / output

$20 / $80

Pricing note: No cached-input discount. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Responses API only, and some requests may take several minutes.

Context

1.04M

Max output

32.7K

Input / output

$2 / $8

Pricing note: OpenAI's Batch table lists no cached-input rate for this model. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Fine-tuned versions follow the same January 14, 2027 shutdown as the base model.

Context

1.04M

Max output

32.7K

Input / output

$0.40 / $1.60

Pricing note: OpenAI's Batch table lists no cached-input rate for this model. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Fine-tuned versions follow the same January 14, 2027 shutdown as the base model.

Context

128K

Max output

16.3K

Input / output

$2.50 / $10

Pricing note: Rates for the gpt-4o alias as listed on October 25, 2026; the pricing page does not say when they took effect. OpenAI's Batch table lists no cached-input rate for gpt-4o.

Caveat: OpenAI deprecated gpt-4o on October 14, 2026, with shutdown on January 14, 2027. The older gpt-4o-2024-05-13 snapshot was shut down on October 23, 2026.

Context

128K

Max output

16.3K

Input / output

$0.15 / $0.60

Pricing note: OpenAI's Batch table lists no cached-input rate for this model. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Fine-tuned versions follow the same January 14, 2027 shutdown as the base model.

Context

400K

Max output

128K

Input / output

$1.25 / $10

Pricing note: Rates as listed on October 25, 2026; the provider does not say when they took effect.

Context

400K

Max output

128K

Input / output

$0.20 / $1.25

Caveat: Deprecated on October 1, 2026; OpenAI will remove it from the API on April 1, 2027. No published launch date; the dated snapshot id suggests March 17, 2026.

Context

1.04M

Max output

65.5K

Input / output

$0.25 / $1.50

Pricing note: Paid-tier text/image/video input price; audio input is billed at $0.50 per 1M. Context caching adds $1.00 per 1M tokens per hour of storage.

Caveat: Google lists May 7, 2027 as the earliest possible shutdown and names Gemini 3.5 Flash-Lite as the replacement.

Context

1M

Max output

Not published

Input / output

Varies

Caveat: The current lifecycle documentation marks Mythos Preview deprecated and directs users to Mythos 5. It does not provide a retirement date; do not treat the former July date as a current commitment.

Context

1.04M

Max output

65.5K

Input / output

Varies

Caveat: Deprecated October 8, 2026; requests now go to Gemini 3.6 Flash, and Google's pricing page no longer lists 3.5 Flash.

Context

1.04M

Max output

65.5K

Input / output

Varies

Caveat: Deprecated October 8, 2026; requests now go to Gemini 3.8 Flash, and Google's pricing page no longer lists 3.7 Flash.

Context

1.04M

Max output

65.5K

Input / output

$2 / $12

Pricing note: Paid-tier price for prompts up to 200K tokens. Longer prompts use higher input and output rates.

Caveat: Preview models can change or retire on shorter notice. gemini-pro-latest is a rolling alias, not a pinned release.

Context

1M

Max output

Not published

Input / output

$1.36 / $4.18

Pricing note: Standard (original) Mistral API rates; batch cached input is $0.07. Mistral's changelog announces launch pricing of 50% off for 2 weeks, shown on the pricing page as a temporary sale ($0.68 input, $0.07 cached input, $2.09 output; batch $0.34/$0.035/$1.045). Mistral publishes no exact end date for the sale, so planning uses the standard rates.

Caveat: Public Preview: Mistral allows silent updates and does not guarantee general availability, and the model card lists the weights and license as coming soon (Mistral says weights ship by the end of October 2026). Mistral publishes no maximum output limit, and mistral-large-latest still points to Mistral Large 3.

Context

1.04M

Max output

Not published

Input / output

$4 / $20

Pricing note: Standard mode; Contributor mode is not available for Muse Wonder on Model API. No long-context premium. Web search grounding adds $2.50 per 1,000 queries.

Caveat: In preview. Muse Wonder Full, the higher-capability variant, is available only in Muse Code with a subscription. Meta publishes no release date or maximum output limit, and the model is not available in South Korea.

Context

300K

Max output

10K

Input / output

Varies

Caveat: Amazon Nova version 1, served through Amazon Bedrock (the alias is the US cross-region inference profile). AWS says Amazon Nova 2 is now available; Nova 2 was not reviewed. Bedrock prices were not reviewed.

Context

128K

Max output

10K

Input / output

Varies

Caveat: Amazon Nova version 1, served through Amazon Bedrock (the alias is the US cross-region inference profile). AWS says Amazon Nova 2 is now available; Nova 2 was not reviewed. Bedrock prices were not reviewed.

Context

1M

Max output

10K

Input / output

Varies

Caveat: Amazon Nova version 1, served through Amazon Bedrock (the alias is the US cross-region inference profile). AWS says Amazon Nova 2 is now available; Nova 2 was not reviewed. Bedrock prices were not reviewed.

Context

300K

Max output

10K

Input / output

Varies

Caveat: Amazon Nova version 1, served through Amazon Bedrock (the alias is the US cross-region inference profile). AWS says Amazon Nova 2 is now available; Nova 2 was not reviewed. Bedrock prices were not reviewed.

Context

1M

Max output

128K

Input / output

$10 / $50

Pricing note: Base Claude API price; regional and partner-platform charges can differ.

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; its model page labels it legacy and recommends Claude Fable 5.1. US export controls forced a global suspension of Fable 5 and Mythos 5 from June 12 to June 30, 2026; access was restored on July 1, 2026.

Context

1M

Max output

128K

Input / output

$10 / $50

Pricing note: Base Claude API price. Cache reads cost 0.025x the base input price on this model. US-only inference (inference_geo) costs 1.1x, and partner-platform prices can differ.

Caveat: Anthropic requires 30-day data retention for this model; it is not available under zero data retention unless Anthropic expressly authorizes it. Forced tool use (tool_choice any or tool) returns a 400 error.

Context

200K

Max output

64K

Input / output

$1 / $5

Pricing note: Base Claude API price; regional and partner-platform charges can differ.

Caveat: The dated API ID is pinned; claude-haiku-4-5 is a convenience alias. Anthropic's not-sooner-than date of October 15, 2026 has passed, and on October 25, 2026 its lifecycle table still listed the model as active with no deprecation notice. Its model page labels it legacy and recommends Claude Haiku 5.5.

Context

1M

Max output

128K

Input / output

$0.10 / $0.50

Pricing note: Base Claude API price for prompts up to 100,000 tokens. US-only inference (inference_geo) costs 1.1x, and partner-platform prices can differ.

Caveat: Unlike other current Claude models, Haiku 5.5 is priced by prompt length: prompts over 100,000 tokens cost $0.50 input and $2.50 output per million tokens. Manual extended thinking (budget_tokens) returns a 400 error.

Context

1M

Max output

128K

Input / output

$10 / $50

Pricing note: Anthropic lists the same specs and pricing as Claude Fable 5.

Caveat: Not generally available: Anthropic offers it only to organizations verified through its verification programs, such as the Cyber Verification Program, which Anthropic merged with Project Glasswing into one expanded program on October 6, 2026; Claude Mythos 5.1 is now the current Mythos model. Also covered by the June 12 to June 30, 2026 export-control suspension; access was restored July 1, 2026.

Context

1M

Max output

128K

Input / output

$10 / $50

Pricing note: Anthropic lists the same specifications and pricing as Claude Fable 5.1, including cache reads at 0.025x the base input price.

Caveat: Not generally available: Anthropic offers it only to organizations verified through its verification programs, such as the Cyber Verification Program (whose three access tiers, announced October 6, 2026, each include it) and the Life Sciences Verification Program; at launch it was limited to a set of US organizations. It is the same model as Claude Fable 5.1 with different safeguards, and it carries 30-day data retention unless Anthropic expressly authorizes zero data retention.

Context

200K

Max output

64K

Input / output

$5 / $25

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; the models overview groups it under legacy models.

Context

1M

Max output

128K

Input / output

$5 / $25

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; the models overview groups it under legacy models.

Context

1M

Max output

128K

Input / output

$5 / $25

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; the models overview groups it under legacy models. Fast mode was deprecated June 25, 2026, with removal on July 24, 2026.

Context

1M

Max output

128K

Input / output

$5 / $25

Pricing note: Base Claude API price; regional and partner-platform charges can differ.

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; its model page labels it legacy and recommends Claude Opus 5.5.

Context

1M

Max output

128K

Input / output

$5 / $25

Pricing note: Base Claude API price, the same as Claude Opus 4.8. Fast mode (research preview, Claude API only) costs $10 input and $50 output per million tokens. US-only inference (inference_geo) costs 1.1x, and partner-platform prices can differ.

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; its model page labels it legacy and recommends migrating to Claude Opus 5.5. Disabling thinking is allowed only at effort high or below.

Context

1M

Max output

128K

Input / output

$4 / $20

Pricing note: Base Claude API price. Cache reads cost 0.05x the base input price on this model. Fast mode (out of research preview since October 15, 2026) costs $8 input and $40 output per million tokens. US-only inference (inference_geo) costs 1.1x, and partner-platform prices can differ.

Caveat: Adaptive thinking is always on and cannot be turned off, and forced tool use (tool_choice any or tool) returns a 400 error. The Message Batches API allows up to 300K output tokens with the output-300k-2026-03-24 beta header.

Context

1M

Max output

128K

Input / output

$3 / $15

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; the models overview groups it under legacy models.

Context

1M

Max output

128K

Input / output

$2 / $10

Pricing note: These are standard prices. The previously announced September 1 increase was cancelled.

Caveat: Anthropic's lifecycle table lists this model as active with a not-sooner-than commitment; its model page labels it legacy and recommends Claude Sonnet 5.5.

Context

1M

Max output

128K

Input / output

$2 / $10

Pricing note: Base Claude API price. Cache reads were $0.20 per million tokens at launch on September 28, 2026 and were cut to $0.10 (0.05x the base input price) on October 7, 2026; all other rates are unchanged since launch. US-only inference (inference_geo) costs 1.1x, and partner-platform prices can differ.

Caveat: Forced tool use (tool_choice any or tool) returns a 400 error, and its thinking blocks work only in the account that produced them or a linked account. The Message Batches API allows up to 300K output tokens with the output-300k-2026-03-24 beta header.

Context

128K

Max output

Not published

Input / output

$0.30 / $0.90

Pricing note: Mistral's Batch rates were not reviewed.

Caveat: Mistral's code completion model (FIM). Mistral publishes no maximum output limit.

Context

256K

Max output

8,000

Input / output

Varies

Caveat: Built for tool use, agents, retrieval-augmented generation and multilingual use. Cohere states context and output limits in rounded thousands; prices were not reviewed.

Context

256K

Max output

32K

Input / output

Varies

Caveat: Cohere's first reasoning model, in 23 languages. Cohere states context and output limits in rounded thousands; prices were not reviewed.

Context

128K

Max output

8,000

Input / output

Varies

Caveat: Cohere's first image-input model; it officially supports English, Portuguese, Italian, French, German and Spanish. Cohere states context and output limits in rounded thousands; prices were not reviewed.

Context

128K

Max output

64K

Input / output

Varies

Caveat: Cohere's first mixture-of-experts model, combining vision input, agentic, reasoning and translation capabilities. Cohere states context and output limits in rounded thousands; prices were not reviewed.

Context

1M

Max output

384K

Input / output

$1.32 / $3.96

Pricing note: Peak rates are the list rates. Off-peak input/cache/output: $0.66/$0.022/$1.98 (half the peak rates). Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday, excluding Chinese public holidays; weekends and Chinese public holidays are off-peak all day. The cost planner blends peak and off-peak rates by the share of traffic sent in peak hours (35 of 168 weekly hours when spread evenly).

Caveat: Served by the DeepSeek-V4-Pro-0813 build since August 13, 2026; image input is not supported. DeepSeek's September 10 news post still says this name routes to V4.1-Flash from September 14, but its change log says V4-Pro API service continues after that date with billing unchanged until further notice, and the pricing page still lists it.

Context

1M

Max output

384K

Input / output

$0.30 / $1.20

Pricing note: Peak rates are the list rates. Off-peak input/cache/output: $0.15/$0.003/$0.60 (half the peak rates). Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday, excluding Chinese public holidays; weekends and Chinese public holidays are off-peak all day. The cost planner blends peak and off-peak rates by the share of traffic sent in peak hours (35 of 168 weekly hours when spread evenly).

Caveat: API model ID is deepseek-flash; the retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, served by this model and billed at these rates. Since October 17, 2026 the API serves the DeepSeek-V4.1-Flash-1017 build, with the same model name and prices. Generic dense-transformer memory formulas do not model this architecture accurately.

Context

1.04M

Max output

65.5K

Input / output

$0.30 / $2.50

Caveat: Not deprecated: Google's deprecation table lists no shutdown date and says the 2.5 models will be served until further notice. Since September 18, 2026 Google limits access to users who have actively used them before, and recommends 3.5 Flash-Lite or 3.8 Flash for new projects.

Context

1.04M

Max output

65.5K

Input / output

$0.10 / $0.40

Caveat: Not deprecated: Google's deprecation table lists no shutdown date and says the 2.5 models will be served until further notice. Since September 18, 2026 Google limits access to users who have actively used them before, and recommends 3.5 Flash-Lite or 3.8 Flash for new projects.

Context

1.04M

Max output

65.5K

Input / output

$1.25 / $10

Pricing note: Paid-tier price for prompts up to 200K tokens. Longer prompts use higher input and output rates.

Caveat: Not deprecated: Google's deprecation table lists no shutdown date and says the 2.5 models will be served until further notice. Since September 18, 2026 Google limits access to users who have actively used them before, and recommends 3.5 Flash-Lite or 3.8 Flash for new projects.

Context

1.04M

Max output

65.5K

Input / output

$0.30 / $2.50

Pricing note: Paid-tier text/image/video input price; feature charges are separate. Context caching is paid-tier only (plus $1.00 per 1M tokens per hour of cache storage).

Caveat: The current Gemini API deprecation table has no shutdown date for this model. Model and service limits remain subject to the linked documentation.

Context

1.04M

Max output

65.5K

Input / output

$0.75 / $3.75

Current pricing through Dec 31, 2026; $1.50/$7.50 from Jan 1, 2027.

Pricing note: Google's Gemini 3.7 Flash guide says the 3.7 introductory rate also applies to 3.6 Flash. The first dated listing is the 3.7 Flash model card published August 13, 2026; Google states no separate start date. Launch list price was $1.50 input and $7.50 output. Cache storage and grounding are billed separately.

Caveat: The current Gemini API deprecation table has no shutdown date for this model. Model and service limits remain subject to the linked documentation.

Context

1.04M

Max output

65.5K

Input / output

$0.75 / $3.75

Current pricing through Dec 31, 2026; $1.50/$7.50 from Jan 1, 2027.

Pricing note: Introductory paid-tier Standard rates announced at launch and valid through December 31, 2026; the output price includes thinking tokens. Cache storage ($0.50 per 1M tokens per hour) and grounding are billed separately.

Caveat: Google qualifies the March 2026 knowledge cutoff: some domains may be limited to January 2025. Google warns the model may use more tokens on complex tasks, especially at higher effort levels.

Context

1.04M

Max output

65.5K

Input / output

$2 / $10

Current pricing through Nov 30, 2026; $4/$20 from Dec 1, 2026.

Pricing note: Introductory paid-tier Standard rates through November 30, 2026; there is no free tier and only implicit caching. The output price includes thinking tokens.

Caveat: Google publishes no knowledge cutoff for Gemini 4 Argon. Its 65,536-token output limit is the default for synchronous requests; Google lists up to 1,048,576 output tokens with long decode.

Context

256K

Max output

Not published

Input / output

Varies

Caveat: Open-weight hosting has no universal price or output cap. The first-party Gemini API deployment can have separate service limits.

Context

1M

Max output

128K

Input / output

$1.40 / $4.40

Pricing note: Cached-input storage is free for a limited time.

Caveat: Text input only. Reasoning is always on (low, high or max). Z.ai compares it with other open-source models and Mistral's catalog lists it as an open-weight model; the license was not reviewed.

Context

1M

Max output

128K

Input / output

$0.15 / $0.50

Pricing note: Cached-input storage is free for a limited time.

Caveat: Z.ai's first native multimodal GLM-5 model, described as open-source; it also accepts files. Thinking cannot be turned off. The license was not reviewed.

Context

1M

Max output

128K

Input / output

$0.37 / $1.25

Pricing note: Cached-input storage is free for a limited time.

Caveat: The faster serving tier of GLM-5.3-Flash (Z.ai cites about 200 tokens per second).

Context

400K

Max output

128K

Input / output

$1.75 / $14

Pricing note: Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: OpenAI calls GPT-5.2 its previous flagship and recommends GPT-6 Astra; the gpt-5.2 model ID has no shutdown date on the deprecations page.

Context

1.05M

Max output

128K

Input / output

$2.50 / $15

Caveat: Positioned by OpenAI as the affordable coding and professional-work model. No published launch date; the dated snapshot id suggests March 5, 2026.

Context

1.05M

Max output

128K

Input / output

$30 / $180

Pricing note: No cached-input discount. OpenAI's pricing table prints Batch rates ($15 / $90), but the model page lists the Batch endpoint as not supported, so no Batch rate is recorded. Regional processing endpoints add 10%. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Responses API only, and some requests may take several minutes.

Context

400K

Max output

128K

Input / output

$0.75 / $4.50

Caveat: No published launch date; the dated snapshot id suggests March 17, 2026.

Context

1.05M

Max output

128K

Input / output

$5 / $30

Caveat: Still sold and priced, but no longer shown on OpenAI's models index. OpenAI has not published a launch date; the dated snapshot id suggests April 23, 2026.

Context

1.05M

Max output

128K

Input / output

$30 / $180

Pricing note: No cached-input discount. Regional processing endpoints add 10%.

Caveat: Available through the Responses API and Batch only, not Chat Completions. OpenAI says some requests may take several minutes and recommends background mode.

Context

1.05M

Max output

128K

Input / output

$0.20 / $1.20

Context

1.05M

Max output

128K

Input / output

$4 / $20

Pricing note: Promotional rates are available at least through November 21, 2026. Later rates have not been committed; projections hold these rates constant.

Caveat: The gpt-5.6 alias currently resolves to Sol; aliases are mutable, so pin gpt-5.6-sol when reproducibility matters.

Context

1.05M

Max output

128K

Input / output

$2 / $12

Context

1.05M

Max output

128K

Input / output

$10 / $50

Pricing note: Flex is priced like Batch and Fast mode costs 2x Standard. Since September 29, 2026 an Ultrafast tier (service_tier ultrafast, Responses API) costs $60 input, $6 cached input, $75 cache writes and $300 output per 1M tokens, or $120, $12, $150 and $450 above 272K input tokens.

Caveat: Does not accept the none reasoning effort, custom temperature or top_p values, or log probabilities, and tool calling requires the Responses API (Chat Completions works without tools).

Context

1.05M

Max output

128K

Input / output

$0.10 / $0.50

Pricing note: Flex is priced like Batch, Fast mode costs 2x Standard, and regional processing adds 10 percent where available.

Caveat: On September 25, 2026 OpenAI fixed an image-encoding bug that had degraded image understanding, so rerun image evaluations made before then. Chat Completions supports function calling only with reasoning effort none; use the Responses API for tools with reasoning.

Context

1.05M

Max output

128K

Input / output

$2 / $10

Pricing note: Flex is priced like Batch, Fast mode costs 2x Standard, and regional processing adds 10 percent where available.

Caveat: OpenAI's docs now point to GPT-6.1 Sol as the newer Sol model, though gpt-6-sol has no announced deprecation. On September 25, 2026 OpenAI fixed an image-encoding bug that had degraded image understanding, so rerun image evaluations made before then.

Context

1.05M

Max output

128K

Input / output

$2 / $10

Pricing note: Cached input is billed at 5 percent of the input rate, half of GPT-6 Sol's cached price. Flex is priced like Batch, Fast mode costs 2x Standard, and regional processing adds 10 percent where available.

Caveat: Does not accept the none or minimal reasoning efforts, and tool calling requires the Responses API (Chat Completions works without tools).

Context

131K

Max output

131K

Input / output

Varies

Caveat: Open weights have no universal token price. Hosting cost, quantization, context limits, and throughput depend on the deployment.

Context

131K

Max output

131K

Input / output

Varies

Caveat: Open weights have no universal token price. Hosting cost, quantization, context limits, and throughput depend on the deployment.

Context

1M

Max output

Not published

Input / output

$1.25 / $2.50

Pricing note: Batch rates are computed from xAI's stated 20% Batch API discount, which applies to all token types including cached tokens. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: xAI publishes no release date, maximum output limit or retirement date. The model page lists more beta and experimental aliases; use the explicit model ID.

Context

1M

Max output

Not published

Input / output

$1.25 / $2.50

Pricing note: Batch rates are computed from xAI's stated 20% Batch API discount, which applies to all token types including cached tokens. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: The non-reasoning variant (the model page lists Reasoning: No). xAI publishes no release date, maximum output limit or retirement date.

Context

1M

Max output

Not published

Input / output

$1.25 / $2.50

Pricing note: Batch rates are computed from xAI's stated 20% Batch API discount, which applies to all token types including cached tokens. Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Beta: multiple agents collaborate in parallel on deep research tasks. Lower rate limits than the single-agent model (9 requests per second). xAI publishes no release date, maximum output limit or retirement date.

Context

1M

Max output

Not published

Input / output

$1.25 / $2.50

Caveat: The official model page lists no maximum output limit or committed retirement date. Alias mappings can change; use the explicit model ID.

Context

500K

Max output

Not published

Input / output

$2 / $6

Caveat: The official model page lists no maximum output limit or committed retirement date. Alias mappings can change; use the explicit model ID.

Context

500K

Max output

Not published

Input / output

$2 / $6

Pricing note: xAI lists a 20% Batch API discount for this model (on all token types); batch rates are not yet in this registry. Requests to the US regional endpoint (https://us.api.x.ai/v1) bill at 1.1x these rates, Priority Processing bills at 2x, and server-side tool calls are billed separately.

Caveat: xAI's model page lists no maximum output limit and xAI publishes no retirement date. Batch API requests are supported in us-east-1 and us-west-2 at a 20% discount to standard rates. The models page recommends Grok 4.7, but grok-4.6 keeps its model page and its pricing row.

Context

500K

Max output

Not published

Input / output

$2 / $6

Pricing note: xAI lists a 20% Batch API discount for this model (on all token types); batch rates are not yet in this registry. Requests to the US regional endpoint (https://us.api.x.ai/v1) bill at 1.1x these rates, Priority Processing bills at 2x, and server-side tool calls are billed separately.

Caveat: xAI's model page lists no maximum output limit and xAI publishes no retirement date. Batch API requests are supported in us-east-1 and us-west-2 at a 20% discount to standard rates. Grok 4.7 Fast (the same model at 2x standard rates, or 1.5x for long-context requests) is sold only in Cursor and Grok Build, not on the public xAI API.

Context

256K

Max output

Not published

Input / output

$1 / $2

Caveat: xAI announced grok-build-0.1 on its API in public beta on May 29, 2026 (its May release notes call it early access). The official model page lists no maximum output limit or committed retirement date, and alias mappings can change, so use the explicit model ID.

Context

1.04M

Max output

1.04M

Input / output

$3 / $15

Pricing note: Flat pricing with no long-context tier. Caching is automatic; cache writes bill once at the 5-minute or 1-hour rate.

Caveat: Thinking is always on (reasoning_effort low, high or max; default max). Output defaults to 131,072 tokens and can be raised to 1,048,576. Moonshot now calls K3 its previous-generation flagship; Kimi K3.1 Pro is listed on the same pricing page. The full weights are released; the license was not reviewed.

Context

1M

Max output

Not published

Input / output

Varies

Caveat: Llama is open-weight under Meta's custom community license, not OSI open source. Hosted providers can impose different limits.

Context

10M

Max output

Not published

Input / output

Varies

Caveat: Llama is open-weight under Meta's custom community license, not OSI open source. Hosted providers may expose a smaller context window.

Context

1.04M

Max output

128K

Input / output

$0.14 / $0.28

Pricing note: Xiaomi MiMo API overseas pay-as-you-go rates (domestic rates are billed in RMB). Cache writes are free for a limited time, and web search costs $5 per 1,000 calls.

Caveat: Open weights on Hugging Face as XiaomiMiMo/MiMo-V2.6-Flash-RL (with a MOPD checkpoint alongside), and served as mimo-v2.6-flash on Xiaomi's MiMo API, which lists a 1M context window (1,048,576 tokens in the weights' config) and 128K maximum output.

Context

1.04M

Max output

128K

Input / output

$0.43 / $0.87

Pricing note: Xiaomi MiMo API overseas pay-as-you-go rates (domestic rates are billed in RMB). Cache writes are free for a limited time, and web search costs $5 per 1,000 calls.

Caveat: Open weights on Hugging Face as XiaomiMiMo/MiMo-V2.6-Pro-RL (with a MOPD checkpoint alongside), and served as mimo-v2.6-pro on Xiaomi's MiMo API, which lists a 1M context window (1,048,576 tokens in the weights' config) and 128K maximum output.

Context

1M

Max output

Not published

Input / output

$0.30 / $1.20

Pricing note: MiniMax shows these as a permanent 50% discount on list prices of $0.60 input and $2.40 output. Priority admission costs 1.5x.

Caveat: Thinking is off by default and can be turned on. MiniMax publishes no maximum output limit; its self-hosting guide calls the open-weight deployment experimental.

Context

1M

Max output

Not published

Input / output

$0.15 / $0.60

Pricing note: Rates as listed on October 25, 2026; the provider does not say when they took effect.

Caveat: Always thinks (thinking depth low to max, default max). MiniMax publishes no maximum output limit.

Context

256K

Max output

Not published

Input / output

$0.20 / $0.20

Pricing note: Mistral's Batch rates were not reviewed.

Caveat: Instruct, reasoning and base weights are published under Apache 2.0. Mistral publishes no maximum output limit.

Context

256K

Max output

Not published

Input / output

$0.10 / $0.10

Pricing note: Mistral's Batch rates were not reviewed.

Caveat: Instruct, reasoning and base weights are published under Apache 2.0. Mistral publishes no maximum output limit.

Context

256K

Max output

Not published

Input / output

$0.15 / $0.15

Pricing note: Mistral's Batch rates were not reviewed.

Caveat: Instruct, reasoning and base weights are published under Apache 2.0. Mistral publishes no maximum output limit.

Context

256K

Max output

Not published

Input / output

$0.50 / $1.50

Pricing note: First-party Mistral API price. Cached-input and batch rates are published directly on Mistral's pricing page (Standard and Batch views).

Caveat: mistral-large-latest is a rolling alias that still points to Mistral Large 3 (mistral-large-2512), not to the Mistral Large 4 public preview. The published model card does not specify a separate maximum-output limit.

Context

256K

Max output

Not published

Input / output

$1.50 / $7.50

Pricing note: First-party Mistral API price. Cached-input and batch rates are published directly on Mistral's pricing page (Standard and Batch views).

Caveat: Mistral publishes dateless IDs only (mistral-medium-3-5 plus the mistral-medium-3 and mistral-medium-latest aliases); the release date and license wording come from the model card alone.

Context

256K

Max output

Not published

Input / output

$0.15 / $0.60

Pricing note: First-party Mistral API price; self-hosting cost is deployment-specific. Cached-input and batch rates are published directly on Mistral's pricing page (Standard and Batch views).

Caveat: mistral-small-latest is a rolling alias. The published model card does not specify a separate maximum-output limit.

Context

1.04M

Max output

Not published

Input / output

$1.25 / $4.25

Pricing note: Standard tier rates as observed on October 8, 2026 (Meta's pricing page does not date them); prompts and completions are not used to train Meta models. No long-context premium; built-in web search grounding adds $2.50 per 1,000 queries. The muse-spark-1.3-contributor variant bills $0.10/$0.002/$0.20 in exchange for training use.

Caveat: Meta lists audio input but says audio understanding in Muse Spark 1.3 is not fully supported and may be degraded (it points audio work to Muse Spark 1.2 or Muse Voice Transcribe). Meta does not publish a maximum output limit, knowledge cutoff, or parameter count, and the max reasoning level is Standard-tier only.

Context

1.04M

Max output

Not published

Input / output

$0.10 / $0.20

Pricing note: Contributor tier rates as observed on October 8, 2026 (Meta's pricing page does not date them): discounted in exchange for permission to use prompts and completions to train future Meta models. The same model on the Standard tier (muse-spark-1.3) costs $1.25/$0.15/$4.25. Built-in web search grounding adds $2.50 per 1,000 queries.

Caveat: Training-eligible tier of Muse Spark 1.3: Meta may use your prompts and completions to train future Meta models, the max reasoning level is not available, and the rate limit is 100 requests and 3,000,000 tokens per minute. Meta does not date this tier's launch, and it says audio understanding in Muse Spark 1.3 is not fully supported.

Context

1.04M

Max output

Not published

Input / output

Varies

Caveat: Not served on Meta's Model API: self-host the BF16 or NVFP4 weights or call a hosted inference provider. Meta lists at least 700GB of GPU memory for NVFP4 and 2TB for BF16. The license is Meta's custom Muse Model License, not an OSI license.

Context

1M

Max output

Not published

Input / output

Varies

Caveat: Alibaba recommends it for balanced performance and cost. Model Studio states a 1M-token context window; the exact input and output limits and the prices were not reviewed.

Context

1M

Max output

Not published

Input / output

Varies

Caveat: Alibaba's lower-cost option; Model Studio lists it under text generation, not image and video understanding. Model Studio states a 1M-token context window; the exact input and output limits and the prices were not reviewed.

Context

1M

Max output

Not published

Input / output

Varies

Caveat: Alibaba recommends it for the strongest reasoning. Model Studio states a 1M-token context window; the exact input and output limits and the prices were not reviewed.

Context

16.3K

Max output

4,096

Input / output

$0.50 / $1.50

Pricing note: OpenAI's pricing table still prints these rates, but the model was shut down on October 23, 2026.

Caveat: OpenAI shut down access to this model on October 23, 2026; requests to it no longer work.

Context

8,192

Max output

8,192

Input / output

$30 / $60

Pricing note: OpenAI's pricing table still prints these rates, but the model was shut down on October 23, 2026.

Caveat: OpenAI shut down access to this model on October 23, 2026; requests to it no longer work.

Context

128K

Max output

4,096

Input / output

$10 / $30

Pricing note: OpenAI's pricing table still prints these rates, but the model was shut down on October 23, 2026.

Caveat: OpenAI shut down access to this model on October 23, 2026; requests to it no longer work.

Context

1.04M

Max output

32.7K

Input / output

Varies

Caveat: OpenAI shut down access to this model on October 23, 2026, along with its fine-tuned versions.

Context

Not published

Max output

Not published

Input / output

$5 / $15

Pricing note: OpenAI's pricing table still prints these rates, but the model was shut down on October 23, 2026.

Caveat: Shut down October 23, 2026. The gpt-4o alias (gpt-4o-2024-08-06, gpt-4o-2024-11-20) was itself deprecated on October 14, 2026 and shuts down January 14, 2027.