# ModelRumors.ai: Complete Frontier AI Reference Corpus (llms-full.txt) > Updated 2026-11-06. Facts checked 2026-11-05 against primary lab documentation and official leaderboards. > Attribution: Cite "ModelRumors.ai (https://modelrumors.ai)" and preserve source URLs. ## 1. Frontier Flagship Verdict & Release Cadence by Lab ### OpenAI (Frontier Flagship: GPT-6 Astra) - **Verdict**: ADOPT NOW · FRESH RELEASE (11 days since GPT-6.1 Terra on 2026-10-26; typical gap ~23 days across recent gaps [56, 19, 7, 27]). - **Rough next-release window at recent pace**: Now to Dec 21, 2026 (Rough estimate from 4 gaps ranging 7 to 56 days. Labs do not follow schedules.) - **Next announced model**: Not announced - **Current lineup**: GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra - **Summary**: OpenAI last shipped GPT-6.1 Terra on Oct 26, 2026, 11 days ago. Its last 4 release gaps were 56, 19, 7, 27 days, so a typical gap is about 23 days. No next model announced as of Nov 5, 2026. - **Dated 2026 release timeline**: - 2026-03-05: GPT-5.4 (API release, counted in gap median) - https://developers.openai.com/api/docs/changelog - 2026-04-24: GPT-5.5 (API release, counted in gap median) - https://openai.com/index/introducing-gpt-5-5/ - 2026-07-09: GPT-5.6 (API release, counted in gap median) - https://openai.com/index/gpt-5-6/ - 2026-09-03: GPT-6 Astra (API release, counted in gap median) - https://openai.com/index/gpt-6-astra/ - 2026-09-22: GPT-6 Sol and Luna (API release, counted in gap median) - https://openai.com/index/introducing-gpt-6-sol-and-luna/ - 2026-09-29: GPT-6.1 Sol (API release, counted in gap median) - https://openai.com/index/introducing-gpt-6-1-sol/ - 2026-10-26: GPT-6.1 Terra (API release, counted in gap median) - https://openai.com/index/introducing-gpt-6-1-terra/ - 2026-10-30: GPT-6 Rosalind (trusted access only, not counted in gap median) - https://deploymentsafety.openai.com/gpt-rosalind-6 - **Official statements**: - [Sep 29, 2026] OpenAI cancelled GPT-6.1 Astra in September and said other models are coming soon. No next model has been announced. (https://techcrunch.com/2026/09/29/openai-launches-gpt-6-1-sol-says-it-nearly-matches-gpt-6-astra-and-costs-less/) - **Unconfirmed rumors (labeled)**: - [The Information, Nov 4, 2026] New OpenAI models, including next-generation Astra and Sol, could arrive as soon as Friday, Nov 6. The report says timelines could change. (https://www.theinformation.com/articles/security-companies-trying-get-ai-agents-control) ### Anthropic (Frontier Flagship: Claude Opus 5.5) - **Verdict**: PAST TYPICAL GAP (30 days since Claude Haiku 5.5 on 2026-10-07; typical gap ~15 days across recent gaps [39, 21, 6, 9]). - **Rough next-release window at recent pace**: Now to Nov 15, 2026 (Rough estimate from 4 gaps ranging 6 to 39 days. Labs do not follow schedules.) - **Next announced model**: Not announced - **Current lineup**: Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5, Claude Fable 5.1 - **Summary**: Anthropic last shipped Claude Haiku 5.5 on Oct 7, 2026, 30 days ago. Its last 4 release gaps were 39, 21, 6, 9 days, so a typical gap is about 15 days. No next model announced as of Nov 5, 2026. - **Dated 2026 release timeline**: - 2026-02-05: Claude Opus 4.6 (API release, counted in gap median) - https://platform.claude.com/docs/en/release-notes/overview - 2026-02-17: Claude Sonnet 4.6 (API release, counted in gap median) - https://platform.claude.com/docs/en/release-notes/overview - 2026-04-16: Claude Opus 4.7 (API release, counted in gap median) - https://platform.claude.com/docs/en/release-notes/overview - 2026-05-28: Claude Opus 4.8 (API release, counted in gap median) - https://platform.claude.com/docs/en/release-notes/overview - 2026-06-09: Claude Fable 5 (API release, counted in gap median) - https://www.anthropic.com/news/claude-fable-5-mythos-5 - 2026-06-30: Claude Sonnet 5 (API release, counted in gap median) - https://www.anthropic.com/news/claude-sonnet-5 - 2026-07-24: Claude Opus 5 (API release, counted in gap median) - https://www.anthropic.com/news/claude-opus-5 - 2026-09-01: Claude Fable 5.1 (API release, counted in gap median) - https://techcrunch.com/2026/09/01/anthropics-new-fable-release-is-cheaper-less-restrictive/ - 2026-09-22: Claude Opus 5.5 (API release, counted in gap median) - https://www.anthropic.com/claude-opus-5-5 - 2026-09-28: Claude Sonnet 5.5 (API release, counted in gap median) - https://www.anthropic.com/claude-sonnet-5-5 - 2026-10-07: Claude Haiku 5.5 (API release, counted in gap median) - https://www.anthropic.com/claude-haiku-5-5 ### Google DeepMind (Frontier Flagship: Gemini 4 Argon) - **Verdict**: MID-CYCLE (24 days since Gemini 4 Argon on 2026-10-13; typical gap ~32 days across recent gaps [63, 23, 20, 41]). - **Rough next-release window at recent pace**: Now to Dec 15, 2026 (Rough estimate from 4 gaps ranging 20 to 63 days. Labs do not follow schedules.) - **Next announced model**: Gemini 4 Carbon - **Current lineup**: Gemini 4 Argon, Gemini 3.8 Flash - **Summary**: Google DeepMind last shipped Gemini 4 Argon on Oct 13, 2026, 24 days ago. Its last 4 release gaps were 63, 23, 20, 41 days, so a typical gap is about 32 days. Next announced model: Gemini 4 Carbon. - **Dated 2026 release timeline**: - 2026-02-19: Gemini 3.1 Pro (public API preview, counted in gap median) - https://ai.google.dev/gemini-api/docs/changelog - 2026-05-19: Gemini 3.5 Flash (API release, counted in gap median) - https://ai.google.dev/gemini-api/docs/changelog - 2026-07-21: Gemini 3.6 Flash (API release, counted in gap median) - https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/ - 2026-08-13: Gemini 3.7 Flash (API release, counted in gap median) - https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/ - 2026-09-02: Gemini 3.8 Flash (API release, counted in gap median) - https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/ - 2026-09-30: Gemini 4 Argon (trusted access only, not counted in gap median) - https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/ - 2026-10-13: Gemini 4 Argon (API release, counted in gap median) - https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/ - 2026-10-28: Gemini 4 Carbon (trusted access only, not counted in gap median) - https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/ - **Official statements**: - [Oct 28, 2026] Gemini 4 Carbon went to a subset of cyber defenders on Oct 28. Google says it will reach developers, businesses and consumers soon, after more guardrail work, at the same $2 / $10 introductory price as Argon. It is not in the API yet. (https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/) - [Nov 5, 2026] Price change ahead: Gemini 4 Argon's introductory $2 / $10 rate ends Nov 30, 2026. It becomes $4 / $20 per 1M tokens on Dec 1. (https://ai.google.dev/gemini-api/docs/pricing) ### xAI (Frontier Flagship: Grok 4.7) - **Verdict**: PAST TYPICAL GAP (46 days since Grok 4.7 on 2026-09-21; typical gap ~46 days across recent gaps [51, 69, 35, 40]). - **Rough next-release window at recent pace**: Now to Nov 29, 2026 (Rough estimate from 4 gaps ranging 35 to 69 days. Labs do not follow schedules.) - **Next announced model**: Grok 5 - **Current lineup**: Grok 4.7 - **Summary**: xAI last shipped Grok 4.7 on Sep 21, 2026, 46 days ago. Its last 4 release gaps were 51, 69, 35, 40 days, so a typical gap is about 46 days. Next announced model: Grok 5. - **Dated 2026 release timeline**: - 2026-03-10: Grok 4.20 (API release, counted in gap median) - https://docs.x.ai/developers/release-notes - 2026-04-30: Grok 4.3 (API release, counted in gap median) - https://artificialanalysis.ai/models/grok-4-3 - 2026-07-08: Grok 4.5 (API release, counted in gap median) - https://techcrunch.com/2026/07/08/spacexai-releases-grok-4-5-which-elon-describes-as-an-opus-class-model/ - 2026-08-12: Grok 4.6 (API release, counted in gap median) - https://x.ai/news/grok-4-6 - 2026-09-21: Grok 4.7 (API release, counted in gap median) - https://x.ai/news/grok-4-7 - **Official statements**: - [Aug 4, 2026] On SpaceX's Q2 earnings call, Elon Musk said Grok 5 should be out before the end of 2026. On the same call he promised Grok 4.7 in three to four weeks; it shipped 48 days later. (https://s21.q4cdn.com/184289198/files/doc_financials/2026/q2/SpaceX-Q2-2026-Earnings_Transcript-FINAL.pdf) ### DeepSeek (Frontier Flagship: DeepSeek V4.1-Flash) - **Verdict**: MID-CYCLE (20 days since DeepSeek V4.1-Flash-1017 on 2026-10-17; typical gap ~32 days across recent gaps [98, 13, 28, 37]). - **Rough next-release window at recent pace**: Now to Jan 23, 2027 (Rough estimate from 4 gaps ranging 13 to 98 days. Labs do not follow schedules.) - **Next announced model**: DeepSeek V4.1-Pro - **Current lineup**: DeepSeek V4.1-Flash, DeepSeek V4-Pro - **Summary**: DeepSeek last shipped DeepSeek V4.1-Flash-1017 on Oct 17, 2026, 20 days ago. Its last 4 release gaps were 98, 13, 28, 37 days, so a typical gap is about 32 days. Next announced model: DeepSeek V4.1-Pro. - **Dated 2026 release timeline**: - 2026-04-24: DeepSeek V4 Preview (V4-Pro, V4-Flash) (open weights, counted in gap median) - https://api-docs.deepseek.com/news/news260424 - 2026-07-31: DeepSeek V4-Flash-0731 (API release, counted in gap median) - https://api-docs.deepseek.com/updates - 2026-08-13: DeepSeek V4-Pro-0813 (open weights, counted in gap median) - https://api-docs.deepseek.com/news/news260813 - 2026-09-10: DeepSeek V4.1-Flash (open weights, counted in gap median) - https://api-docs.deepseek.com/news/news260910 - 2026-10-17: DeepSeek V4.1-Flash-1017 (open weights, counted in gap median) - https://api-docs.deepseek.com/news/news261017 - **Official statements**: - [Sep 10, 2026] DeepSeek says V4-Pro API service continues until V4.1-Pro launches. It has given no date. (https://api-docs.deepseek.com/news/news260910) ### Meta (Frontier Flagship: Muse Wonder) - **Verdict**: PAST TYPICAL GAP (22 days since Muse Spark Open on 2026-10-15; typical gap ~14 days across recent gaps [5, 23, 40, 3]). - **Rough next-release window at recent pace**: Now to Nov 24, 2026 (Rough estimate from 4 gaps ranging 3 to 40 days. Labs do not follow schedules.) - **Next announced model**: Not announced - **Current lineup**: Muse Wonder, Muse Spark 1.3, Muse Spark Open, Muse Glimmer - **Summary**: Meta last shipped Muse Spark Open on Oct 15, 2026, 22 days ago. Its last 4 release gaps were 5, 23, 40, 3 days, so a typical gap is about 14 days. No next model announced as of Nov 5, 2026. - **Dated 2026 release timeline**: - 2026-04-08: Muse Spark (app only, not counted in gap median) - https://ai.meta.com/blog/introducing-muse-spark-msl/ - 2026-07-09: Muse Spark 1.1 (public API preview, counted in gap median) - https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/ - 2026-08-05: Muse Spark 1.2 (public API preview, counted in gap median) - https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2 - 2026-08-10: Muse Glimmer 30B (open weights, counted in gap median) - https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model - 2026-09-02: Muse Spark 1.3 (public API preview, counted in gap median) - https://research.meta.ai/blog/introducing-muse-spark-1-3 - 2026-10-12: Muse Wonder (public API preview, counted in gap median) - https://about.fb.com/news/2026/10/introducing-muse-wonder-models/ - 2026-10-15: Muse Spark Open (open weights, counted in gap median) - https://research.meta.ai/blog/muse-spark-open ## 2. Complete Model Specifications, API Strings, SDK Snippets & Pricing Edge Cases ### Gemini 4 Argon (Google DeepMind) [LAB FRONTIER FLAGSHIP] - **API Model ID**: `gemini-4-argon` (checked 2026-11-05 on https://aistudio.google.com/docs/thinking) - **Release Date**: 2026-10-13 (https://ai.google.dev/gemini-api/docs/models/gemini-4-argon) - **Arena Elo**: 1520 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 96.4% (Epoch AI, https://epoch.ai/models/gemini-4-argon) - **Additional Benchmarks**: DeepSWE v1.1: 77.9% (Google, https://deepmind.google/models/model-cards/gemini-4-argon/); AA Intelligence Index v5.0.1: 50 (High) (Artificial Analysis, https://artificialanalysis.ai/models/gemini-4-argon) - **Standard Token Price**: $2 / $10 per 1M input/output tokens ($4.00/1M blended at 3:1) (https://ai.google.dev/gemini-api/docs/pricing) - **Scheduled Price Step-Up**: Steps up on 2026-12-01 to $4.00 / $20.00 per 1M input/output tokens. - **Long Prompts**: Same rate at any prompt length. (https://ai.google.dev/gemini-api/docs/pricing) - **Pricing Note**: Introductory price until Nov 30, 2026; $4 / $20 from Dec 1. No higher rate for long prompts. - **Context Window**: 1M | **Throughput**: 74 tok/s | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high` (Default: `medium`; effort used for headline scores: `high`) - **Effort API Parameter**: `thinking_config={"thinking_level": "low" | "medium" | "high"}` - **Python SDK Snippet**: `client.models.generate_content(model="gemini-4-argon", config={"thinking_config": {"thinking_level": "medium"}}, contents=prompt)` - **Best For**: Highest GPQA Diamond score we track (96.4%, Epoch AI) and the top Arena Text score among models you can use today (1520). (https://epoch.ai/models/gemini-4-argon) ### Claude Opus 5.5 (Anthropic) [LAB FRONTIER FLAGSHIP] - **API Model ID**: `claude-opus-5-5` (checked 2026-11-05 on https://platform.claude.com/docs/en/build-with-claude/effort) - **Release Date**: 2026-09-22 (https://www.anthropic.com/claude-opus-5-5) - **Arena Elo**: 1509 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 90.6% (Epoch AI, max effort, https://epoch.ai/benchmarks/gpqa-diamond) - **Additional Benchmarks**: Arena WebDev: 1810 (#1) (Arena, https://arena.ai/leaderboard/webdev); SWE-bench Pro: 89.9% (Anthropic system card, https://www.anthropic.com/document/claude-opus-5-5-system-card); Terminal-Bench 4.0: 66.4% (xhigh) (Anthropic, https://www.anthropic.com/claude-opus-5-5); GDPval-AA v2.1: 1846 (Anthropic, https://www.anthropic.com/claude-opus-5-5); CursorBench 4.0: 57.8% at max effort, 52.5% at default (Anthropic, https://www.anthropic.com/claude-opus-5-5); AA Intelligence Index v5.0.1: 59 (Artificial Analysis, https://artificialanalysis.ai/models/claude-opus-5-5) - **Standard Token Price**: $4 / $20 per 1M input/output tokens ($8.00/1M blended at 3:1) (https://platform.claude.com/docs/en/about-claude/pricing) - **Long Prompts**: Same rate at any prompt length. (https://platform.claude.com/docs/en/about-claude/pricing) - **Pricing Note**: Same rate across the full 1M context window. - **Context Window**: 1M | **Throughput**: 110 tok/s | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high / xhigh / max` (Default: `medium`; effort used for headline scores: `max`) - **Effort API Parameter**: `output_config={"effort": "low" | "medium" | "high" | "xhigh" | "max"}` - **Python SDK Snippet**: `client.messages.create(model="claude-opus-5-5", max_tokens=16000, output_config={"effort": "medium"}, messages=[{"role": "user", "content": prompt}])` - **Best For**: Anthropic suggests starting with Opus 5.5 for most workloads. #1 on Arena WebDev (1810); 66.4% on Terminal-Bench 4.0 at xhigh effort (Anthropic). (https://platform.claude.com/docs/en/models/fable-5-1/overview) ### Muse Wonder (Meta) [LAB FRONTIER FLAGSHIP] - **API Model ID**: `muse-wonder-1.0` (checked 2026-11-05 on https://dev.meta.ai/docs/reasoning) - **Release Date**: 2026-10-12 (https://about.fb.com/news/2026/10/introducing-muse-wonder-models/) - **Arena Elo**: 1507 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: not published - **Additional Benchmarks**: Arena Text: 1507 (#4, Max effort) (Arena, https://arena.ai/leaderboard/text) - **Standard Token Price**: $4 / $20 per 1M input/output tokens ($8.00/1M blended at 3:1) (https://dev.meta.ai/docs/pricing-rate-limits) - **Long Prompts**: Long-prompt pricing not published in our sources. - **Pricing Note**: Meta Model API, in preview. - **Context Window**: not published | **Throughput**: not published | **Open Weights**: No - **Reasoning Effort Tiers**: `minimal / low / medium / high / xhigh / max` (Default: `not stated`; effort used for headline scores: `max`) - **Effort API Parameter**: `reasoning_effort="minimal" | "low" | "medium" | "high" | "xhigh" | "max" (Chat Completions)` - **Python SDK Snippet**: `client.chat.completions.create(model="muse-wonder-1.0", reasoning_effort="high", messages=[{"role": "user", "content": prompt}])` - **Best For**: Meta's highest-ranked model on Arena Text (1507, #4 at Max effort). The API is in preview. (https://arena.ai/leaderboard/text) ### Claude Fable 5.1 (Anthropic) - **API Model ID**: `claude-fable-5-1` (checked 2026-11-05 on https://platform.claude.com/docs/en/build-with-claude/effort) - **Release Date**: 2026-09-01 (https://techcrunch.com/2026/09/01/anthropics-new-fable-release-is-cheaper-less-restrictive/) - **Arena Elo**: 1504 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: not published - **Additional Benchmarks**: Terminal-Bench 4.0: 55.8% (Anthropic, https://www.anthropic.com/claude-opus-5-5); GDPval-AA v2.1: 1735 (Anthropic, https://www.anthropic.com/claude-opus-5-5); AA Intelligence Index v5.0.1: 50 (Artificial Analysis, https://artificialanalysis.ai/models/claude-fable-5-1) - **Standard Token Price**: $10 / $50 per 1M input/output tokens ($20.00/1M blended at 3:1) (https://platform.claude.com/docs/en/about-claude/pricing) - **Long Prompts**: Same rate at any prompt length. (https://platform.claude.com/docs/en/about-claude/pricing) - **Pricing Note**: Same rate across the full 1M context window. - **Context Window**: 1M | **Throughput**: not published | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high / xhigh / max` (Default: `high`; effort used for headline scores: `not stated`) - **Effort API Parameter**: `output_config={"effort": "low" | "medium" | "high" | "xhigh" | "max"}` - **Python SDK Snippet**: `client.messages.create(model="claude-fable-5-1", max_tokens=16000, output_config={"effort": "high"}, messages=[{"role": "user", "content": prompt}])` - **Best For**: Anthropic's most capable generally available model, for demanding reasoning and long-horizon agentic work (Anthropic). (https://www.anthropic.com/claude/fable) ### Gemini 3.8 Flash (Google DeepMind) - **API Model ID**: `gemini-3.8-flash` (checked 2026-11-05 on https://aistudio.google.com/docs/thinking) - **Release Date**: 2026-09-02 (https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash) - **Arena Elo**: 1496 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 95.4% (Epoch AI, https://epoch.ai/models/gemini-3-8-flash) - **Additional Benchmarks**: DeepSWE v1.1: 73.7% (Google, https://deepmind.google/models/model-cards/gemini-4-argon/) - **Standard Token Price**: $0.75 / $3.75 per 1M input/output tokens ($1.50/1M blended at 3:1) (https://ai.google.dev/gemini-api/docs/pricing) - **Scheduled Price Step-Up**: Steps up on 2027-01-01 to $1.50 / $7.50 per 1M input/output tokens. - **Long Prompts**: Same rate at any prompt length. (https://ai.google.dev/gemini-api/docs/pricing) - **Pricing Note**: Until Dec 31, 2026; $1.50 / $7.50 from Jan 1, 2027. No higher rate for long prompts. - **Context Window**: 1M | **Throughput**: 117 tok/s | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high` (Default: `medium`; effort used for headline scores: `high`) - **Effort API Parameter**: `thinking_config={"thinking_level": "low" | "medium" | "high"}` - **Python SDK Snippet**: `client.models.generate_content(model="gemini-3.8-flash", config={"thinking_config": {"thinking_level": "medium"}}, contents=prompt)` - **Best For**: 95.4% on GPQA Diamond (Epoch AI) and 117 tok/s output (Artificial Analysis) at a Flash-tier price. (https://epoch.ai/models/gemini-3-8-flash) ### GPT-6.1 Sol (OpenAI) - **API Model ID**: `gpt-6.1-sol` (checked 2026-11-05 on https://developers.openai.com/api/docs/models/gpt-6.1-sol) - **Release Date**: 2026-09-29 (https://openai.com/index/introducing-gpt-6-1-sol/) - **Arena Elo**: 1481 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: not published - **Additional Benchmarks**: AA Intelligence Index v5.0.1: 55 (Max) (Artificial Analysis, https://artificialanalysis.ai/models/gpt-6-1-sol) - **Standard Token Price**: $2 / $10 per 1M input/output tokens ($4.00/1M blended at 3:1) (https://developers.openai.com/api/docs/pricing) - **Long Prompts (above 272K input tokens)**: $4.00 / $15.00 per 1M input/output tokens. (https://developers.openai.com/api/docs/pricing) - **Pricing Note**: Prompts up to 272K tokens; $4 / $15 above 272K. OpenAI describes Sol as near-Astra at lower cost. - **Context Window**: 1.05M | **Throughput**: not published | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high / xhigh / max` (Default: `medium`; effort used for headline scores: `max`) - **Effort API Parameter**: `reasoning={"effort": "low" | "medium" | "high" | "xhigh" | "max"}` - **Python SDK Snippet**: `client.responses.create(model="gpt-6.1-sol", reasoning={"effort": "medium"}, input=prompt)` - **Best For**: OpenAI's near-Astra model at one-fifth of Astra's token price, with a 1.05M-token context window. (https://developers.openai.com/api/docs/models/gpt-6.1-sol) ### Claude Sonnet 5.5 (Anthropic) - **API Model ID**: `claude-sonnet-5-5` (checked 2026-11-05 on https://platform.claude.com/docs/en/build-with-claude/effort) - **Release Date**: 2026-09-28 (https://www.anthropic.com/claude-sonnet-5-5) - **Arena Elo**: 1477 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 95.6% (Epoch AI, max effort, https://epoch.ai/benchmarks/gpqa-diamond) - **Additional Benchmarks**: SWE-bench Pro: 81.3% (Anthropic system card, https://www.anthropic.com/document/claude-sonnet-5-5-system-card); AA Intelligence Index v5.0.1: 55 (Artificial Analysis, https://artificialanalysis.ai/models/claude-sonnet-5-5) - **Standard Token Price**: $2 / $10 per 1M input/output tokens ($4.00/1M blended at 3:1) (https://platform.claude.com/docs/en/about-claude/pricing) - **Long Prompts**: Same rate at any prompt length. (https://platform.claude.com/docs/en/about-claude/pricing) - **Pricing Note**: Same rate across the full 1M context window. - **Context Window**: 1M | **Throughput**: 155 tok/s | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high / xhigh / max` (Default: `high`; effort used for headline scores: `max`) - **Effort API Parameter**: `output_config={"effort": "low" | "medium" | "high" | "xhigh" | "max"}` - **Python SDK Snippet**: `client.messages.create(model="claude-sonnet-5-5", max_tokens=16000, output_config={"effort": "high"}, messages=[{"role": "user", "content": prompt}])` - **Best For**: Fast (155 tok/s, Artificial Analysis) and strong on science: 95.6% GPQA Diamond at max effort (Epoch AI), above Opus 5.5's 90.6%. (https://epoch.ai/benchmarks/gpqa-diamond) ### GPT-6 Astra (OpenAI) [LAB FRONTIER FLAGSHIP] - **API Model ID**: `gpt-6-astra` (checked 2026-11-05 on https://developers.openai.com/api/docs/guides/reasoning) - **Release Date**: 2026-09-03 (https://openai.com/index/gpt-6-astra/) - **Arena Elo**: 1476 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 96.0% (OpenAI, https://openai.com/index/gpt-6-astra/) - **Additional Benchmarks**: DeepSWE v1.1: 74.1% (OpenAI, https://openai.com/index/gpt-6-astra/); Terminal-Bench 4.0: 57.9% (OpenAI, https://openai.com/index/gpt-6-astra/); AA Intelligence Index v5.0.1: 57 (Max) (Artificial Analysis, https://artificialanalysis.ai/models/gpt-6-astra) - **Standard Token Price**: $10 / $50 per 1M input/output tokens ($20.00/1M blended at 3:1) (https://developers.openai.com/api/docs/pricing) - **Long Prompts (above 272K input tokens)**: $20.00 / $75.00 per 1M input/output tokens. (https://developers.openai.com/api/docs/pricing) - **Batch API (24h async)**: $5.00 / $25.00 per 1M input/output tokens. - **Pricing Note**: Prompts up to 272K tokens; $20 / $75 above 272K. Batch and Flex: $5 / $25. - **Context Window**: 1.05M | **Throughput**: not published | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high / xhigh / max` (Default: `not stated`; effort used for headline scores: `max`) - **Effort API Parameter**: `reasoning={"effort": "low" | "medium" | "high" | "xhigh" | "max"}` - **Python SDK Snippet**: `client.responses.create(model="gpt-6-astra", reasoning={"effort": "medium"}, input=prompt)` - **Best For**: OpenAI's most capable model: 96.0% GPQA Diamond, 74.1% DeepSWE v1.1 and 57.9% Terminal-Bench 4.0 (OpenAI). (https://openai.com/index/gpt-6-astra/) ### DeepSeek V4.1-Flash (DeepSeek) [LAB FRONTIER FLAGSHIP] - **API Model ID**: `deepseek-flash` (checked 2026-11-05 on https://api-docs.deepseek.com/guides/thinking_mode) - **Release Date**: 2026-09-10 (https://api-docs.deepseek.com/news/news261017) - **Arena Elo**: 1476 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 90.9% (DeepSeek, https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash) - **Additional Benchmarks**: DeepSWE v1.1: 77.0% (Oct 17 update; 74.2% on Sep 10) (DeepSeek, https://api-docs.deepseek.com/news/news261017) - **Standard Token Price**: $0.15 / $0.60 per 1M input/output tokens ($0.26/1M blended at 3:1) (https://api-docs.deepseek.com/quick_start/pricing) - **Peak-Hour Price** (01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, except Chinese public holidays): $0.30 / $1.20 per 1M input/output tokens. Cache-hit input: $0.003 off-peak / $0.006 peak. (https://api-docs.deepseek.com/quick_start/pricing) - **Long Prompts**: Long-prompt pricing not published in our sources. - **Pricing Note**: Off-peak price (cache hit $0.003 / 1M). Peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, weekdays): $0.30 / $1.20 (cache hit $0.006 / 1M). - **Context Window**: 1M | **Throughput**: 174 tok/s | **Open Weights**: Yes - **Reasoning Effort Tiers**: `low / high / max` (Default: `high`; effort used for headline scores: `not stated`) - **Effort API Parameter**: `thinking={"type": "enabled" | "disabled"}, reasoning_effort="low" | "high" | "max" (medium and xhigh map to high)` - **Python SDK Snippet**: `client.chat.completions.create(model="deepseek-flash", reasoning_effort="high", extra_body={"thinking": {"type": "enabled"}}, messages=[{"role": "user", "content": prompt}])` - **Best For**: Fastest output we track (174 tok/s, Artificial Analysis) and the lowest list price ($0.15 / $0.60 off-peak); MIT open weights. (https://artificialanalysis.ai/models/deepseek-v4-1-flash-1017) ### DeepSeek V4-Pro (DeepSeek) - **API Model ID**: `deepseek-v4-pro` (checked 2026-11-05 on https://api-docs.deepseek.com/guides/thinking_mode) - **Release Date**: 2026-08-13 (https://api-docs.deepseek.com/news/news260813) - **Arena Elo**: 1465 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 92.4% (DeepSeek, https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash) - **Additional Benchmarks**: SWE-bench Verified: 80.6% (April V4 preview card, Max) (DeepSeek, https://api-docs.deepseek.com/news/news260424) - **Standard Token Price**: $0.66 / $1.98 per 1M input/output tokens ($0.99/1M blended at 3:1) (https://api-docs.deepseek.com/quick_start/pricing) - **Peak-Hour Price** (01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, except Chinese public holidays): $1.32 / $3.96 per 1M input/output tokens. Cache-hit input: $0.022 off-peak / $0.044 peak. (https://api-docs.deepseek.com/quick_start/pricing) - **Long Prompts**: Long-prompt pricing not published in our sources. - **Pricing Note**: Off-peak price (cache hit $0.022 / 1M). Peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, weekdays): $1.32 / $3.96 (cache hit $0.044 / 1M). API continues until V4.1-Pro launches. - **Context Window**: 1M | **Throughput**: not published | **Open Weights**: Yes - **Reasoning Effort Tiers**: `low / high / max` (Default: `high`; effort used for headline scores: `max`) - **Effort API Parameter**: `thinking={"type": "enabled" | "disabled"}, reasoning_effort="low" | "high" | "max" (medium and xhigh map to high)` - **Python SDK Snippet**: `client.chat.completions.create(model="deepseek-v4-pro", reasoning_effort="high", extra_body={"thinking": {"type": "enabled"}}, messages=[{"role": "user", "content": prompt}])` - **Best For**: Higher GPQA Diamond than V4.1-Flash (92.4% vs 90.9%, DeepSeek); MIT open weights. (https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash) ### Grok 4.7 (xAI) [LAB FRONTIER FLAGSHIP] - **API Model ID**: `grok-4.7` (checked 2026-11-05 on https://docs.x.ai/developers/model-capabilities/text/reasoning) - **Release Date**: 2026-09-21 (https://x.ai/news/grok-4-7) - **Arena Elo**: 1441 (https://arena.ai/leaderboard/text) - **GPQA Diamond**: not published - **Additional Benchmarks**: DeepSWE v1.1: 71.0% (xAI, https://x.ai/news/grok-4-7); CursorBench 4.0: 46.3% (xAI, https://x.ai/news/grok-4-7); AA Intelligence Index v5.0.1: 35 (xhigh) (Artificial Analysis, https://artificialanalysis.ai/models/grok-4-7) - **Standard Token Price**: $2 / $6 per 1M input/output tokens ($3.00/1M blended at 3:1) (https://docs.x.ai/developers/models) - **Long Prompts (above 200K input tokens)**: $4.00 / $12.00 per 1M input/output tokens. (https://docs.x.ai/developers/models) - **Pricing Note**: Prompts under 200K tokens; $4 / $12 above 200K. - **Context Window**: 500K | **Throughput**: 72 tok/s | **Open Weights**: No - **Reasoning Effort Tiers**: `low / medium / high / xhigh` (Default: `high`; effort used for headline scores: `xhigh`) - **Effort API Parameter**: `reasoning_effort="low" | "medium" | "high" | "xhigh"` - **Python SDK Snippet**: `client.chat.completions.create(model="grok-4.7", reasoning_effort="high", messages=[{"role": "user", "content": prompt}])` - **Best For**: xAI's flagship. Reasoning is always on and cannot be turned off; 500K context; 71.0% on DeepSWE v1.1 (xAI). (https://docs.x.ai/developers/model-capabilities/text/reasoning) ### Muse Spark Open (Meta) - **API Model ID**: `meta/muse-spark-open` (checked 2026-11-05 on https://openrouter.ai/meta/muse-spark-open) - **Release Date**: 2026-10-15 (https://research.meta.ai/blog/muse-spark-open) - **Arena Elo**: not ranked (https://arena.ai/leaderboard/text) - **GPQA Diamond**: 89.0% (Meta, https://research.meta.ai/blog/muse-spark-open) - **Additional Benchmarks**: Size: 816B total, 42B active (MoE) (Meta, https://research.meta.ai/blog/muse-spark-open); Output speed: 153 tok/s, varies by host (Artificial Analysis, https://artificialanalysis.ai/models/muse-spark-open) - **Standard Token Price**: $1.25 / $4.25 per 1M input/output tokens ($2.00/1M blended at 3:1) (https://openrouter.ai/meta/muse-spark-open) - **Long Prompts**: Long-prompt pricing not published in our sources. - **Pricing Note**: Hosted price on OpenRouter (ID meta/muse-spark-open) and Together AI (ID meta-models/Muse-Spark-Open-816B-A42B). Open weights under the Muse Model License. - **Context Window**: 1M | **Throughput**: 153 tok/s | **Open Weights**: Yes - **Reasoning Effort Tiers**: `not verified` (Default: `not stated`; effort used for headline scores: `not stated`) - **Effort API Parameter**: `Not verified: reasoning controls depend on the host` - **Python SDK Snippet**: `client.chat.completions.create(model="meta/muse-spark-open", messages=[{"role": "user", "content": prompt}])` - **Best For**: Meta's large open-weights model (816B total, 42B active) with a 1M context window; 89.0% GPQA Diamond (Meta). (https://research.meta.ai/blog/muse-spark-open) ## 3. Reasoning Effort-Level Matchups & Break-Even Analysis ### GPT-6 Astra or GPT-6.1 Sol: how do they compare? [Sol: near-Astra at one-fifth of the price (OpenAI)] - **Left**: GPT-6 Astra at `low to max` ($10 / $50 per 1M, API `reasoning={"effort": "high"}`) - **Right**: GPT-6.1 Sol at `low to max (default medium)` ($2 / $10 per 1M, API `reasoning={"effort": "max"}`) - **Analysis**: OpenAI describes GPT-6.1 Sol as near-Astra in performance at lower cost: $2 / $10 per 1M input/output tokens against $10 / $50 for GPT-6 Astra, for prompts up to 272K tokens. On Arena Text, both rated at Max effort, Sol scores 1481 and Astra 1476, a gap too small to call. Artificial Analysis scores Astra 57 and Sol 55 on its Intelligence Index (Max). OpenAI publishes DeepSWE v1.1 (74.1%) and Terminal-Bench 4.0 (57.9%) for Astra, and no comparable scores for Sol. Reasoning tokens bill as output, so higher effort costs more; the break-even calculator shows how many more output tokens Sol can use before it costs as much as Astra. - **Sources**: https://developers.openai.com/api/docs/models/gpt-6.1-sol, https://openai.com/index/gpt-6-astra/, https://arena.ai/leaderboard/text, https://artificialanalysis.ai/models/gpt-6-astra, https://developers.openai.com/api/docs/pricing ### Which models match GPT-6 Astra on science and coding benchmarks? [Several models match Astra on GPQA Diamond for less] - **Left**: GPT-6 Astra at `max` ($10 / $50 per 1M, API `reasoning={"effort": "max"}`) - **Right**: Other labs at `as each source reports` (see below, API `see each model`) - **Analysis**: GPQA Diamond (graduate science): Gemini 4 Argon 96.4% (Epoch AI), GPT-6 Astra 96.0% (reported by OpenAI), Claude Sonnet 5.5 95.6% (Epoch AI, max effort) and Gemini 3.8 Flash 95.4% (Epoch AI). The evaluators differ, so treat gaps under a point as ties. Coding and agents: OpenAI reports 57.9% on Terminal-Bench 4.0 for Astra; Anthropic reports 66.4% for Claude Opus 5.5 at xhigh effort. Artificial Analysis Intelligence Index v5.0.1: Opus 5.5 59, Astra 57 (Max), Sol 55 (Max), Argon 50 (High). Prices per 1M input/output: Astra $10 / $50, Opus 5.5 $4 / $20, Argon $2 / $10, Sonnet 5.5 $2 / $10, 3.8 Flash $0.75 / $3.75. - **Sources**: https://epoch.ai/benchmarks/gpqa-diamond, https://openai.com/index/gpt-6-astra/, https://www.anthropic.com/claude-opus-5-5, https://artificialanalysis.ai/models/gpt-6-astra ### Claude Opus 5.5 or Claude Sonnet 5.5, and where does Fable 5.1 fit? [Opus 5.5 for coding · Sonnet 5.5 (max) for GPQA science] - **Left**: Claude Opus 5.5 at `medium (default) to xhigh` ($4 / $20 per 1M, API `output_config={"effort": "xhigh"}`) - **Right**: Claude Sonnet 5.5 at `max` ($2 / $10 per 1M, API `output_config={"effort": "max"}`) - **Analysis**: Anthropic suggests starting with Claude Opus 5.5 for most workloads; its API default effort is medium. Opus 5.5 leads Sonnet 5.5 on SWE-bench Pro (89.9% vs 81.3%, Anthropic system cards), ranks #1 on Arena WebDev (1810) and scores 1509 vs 1477 on Arena Text. On GPQA Diamond, run by Epoch AI at max effort, Sonnet 5.5 scores 95.6% against 90.6% for Opus 5.5, streams faster (155 vs 110 tok/s, Artificial Analysis) and costs half as much ($2 / $10 vs $4 / $20). Anthropic calls Claude Fable 5.1 ($10 / $50) its most capable generally available model, for demanding reasoning and long-horizon agentic work; on Terminal-Bench 4.0, Anthropic reports Opus 5.5 at 66.4% and Fable 5.1 at 55.8%. - **Sources**: https://www.anthropic.com/claude-opus-5-5, https://www.anthropic.com/document/claude-opus-5-5-system-card, https://www.anthropic.com/document/claude-sonnet-5-5-system-card, https://epoch.ai/benchmarks/gpqa-diamond, https://platform.claude.com/docs/en/build-with-claude/effort, https://www.anthropic.com/claude/fable, https://platform.claude.com/docs/en/models/fable-5-1/overview ### Gemini 4 Argon or Gemini 3.8 Flash? [3.8 Flash: 99% of Argon's GPQA score at 38% of the price] - **Left**: Gemini 4 Argon at `low to high (default medium)` ($2 / $10 per 1M, API `thinking_level="high"`) - **Right**: Gemini 3.8 Flash at `low to high (default medium)` ($0.75 / $3.75 per 1M, API `thinking_level="high"`) - **Analysis**: On Epoch AI's GPQA Diamond, Gemini 3.8 Flash scores 95.4% against 96.4% for Gemini 4 Argon, and Google reports 73.7% against 77.9% on DeepSWE v1.1. 3.8 Flash streams at 117 tok/s against 74 (Artificial Analysis) and costs $0.75 / $3.75 against $2 / $10 per 1M. On Arena Text (High thinking), Argon scores 1520 and 3.8 Flash 1496 (preliminary). Scheduled changes: Gemini 4 Argon becomes $4 / $20 on Dec 1, 2026; Gemini 3.8 Flash becomes $1.5 / $7.5 on Jan 1, 2027. - **Sources**: https://epoch.ai/models/gemini-4-argon, https://epoch.ai/models/gemini-3-8-flash, https://deepmind.google/models/model-cards/gemini-4-argon/, https://ai.google.dev/gemini-api/docs/pricing ### DeepSeek V4.1-Flash or DeepSeek V4-Pro? [V4.1-Flash: same Arena tier, faster and cheaper · V4-Pro: higher GPQA] - **Left**: DeepSeek V4.1-Flash at `low, high (default) or max` ($0.15 / $0.60 off-peak, API `reasoning_effort="max"`) - **Right**: DeepSeek V4-Pro at `low, high (default) or max` ($0.66 / $1.98 off-peak, API `reasoning_effort="max"`) - **Analysis**: On Arena Text, V4.1-Flash (Sep 10 version) scores 1476 ±6 and V4-Pro-0813 scores 1465 ±7, a gap inside the margin of error. V4-Pro scores higher on GPQA Diamond (92.4% vs 90.9%, DeepSeek). The only coding test both have is SWE-bench Verified from the April V4 preview cards, at Max: V4-Pro 80.6%, V4-Flash 79.0%. The Oct 17 V4.1-Flash update reports 77.0% on DeepSWE v1.1. Off-peak prices per 1M: V4.1-Flash $0.15 / $0.60, V4-Pro $0.66 / $1.98, so Flash costs 23% to 30% as much. DeepSeek says V4-Pro API service continues until V4.1-Pro launches. - **Sources**: https://api-docs.deepseek.com/news/news261017, https://api-docs.deepseek.com/news/news260813, https://api-docs.deepseek.com/news/news260424, https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash, https://api-docs.deepseek.com/quick_start/pricing, https://arena.ai/leaderboard/text ## 4. Workload & Use-Case Selection Matrix - **1. Multi-File Agentic Coding, Terminal Tasks & WebDev UI** - Best Overall: Claude Opus 5.5 (xhigh effort) - Best Value: GPT-6.1 Sol (high effort) - Best Open Weights: DeepSeek V4.1-Flash (max) - Evidence: Anthropic reports Claude Opus 5.5 at 66.4% on Terminal-Bench 4.0 (xhigh), 57.8% on CursorBench 4.0 (max effort; 52.5% at its default effort) and 89.9% on SWE-bench Pro, and it ranks #1 on Arena WebDev (1810). OpenAI describes GPT-6.1 Sol as near-Astra at $2 / $10 per 1M. Google reports 77.9% on DeepSWE v1.1 for Gemini 4 Argon ($2 / $10). Among open weights, DeepSeek V4.1-Flash-1017 reports 77.0% on DeepSWE v1.1 at $0.15 / $0.60 off-peak. (Sources: https://www.anthropic.com/claude-opus-5-5, https://www.anthropic.com/document/claude-opus-5-5-system-card, https://developers.openai.com/api/docs/models/gpt-6.1-sol, https://deepmind.google/models/model-cards/gemini-4-argon/, https://api-docs.deepseek.com/news/news261017) - **2. Graduate-Level Science, Quantitative Research & Math (GPQA Diamond)** - Best Overall: Gemini 4 Argon (high thinking) · 96.4% GPQA - Best Value: Gemini 3.8 Flash (high thinking) · 95.4% GPQA at $0.75 / $3.75 - Best Open Weights: DeepSeek V4-Pro (max) · 92.4% GPQA - Evidence: GPQA Diamond: Gemini 4 Argon 96.4% (Epoch AI), GPT-6 Astra 96.0% (reported by OpenAI), Claude Sonnet 5.5 95.6% (Epoch AI, max effort) and Gemini 3.8 Flash 95.4% (Epoch AI). Sonnet 5.5 outscores Opus 5.5 (90.6%) here. Among open weights, DeepSeek reports 92.4% for V4-Pro. (Sources: https://epoch.ai/models/gemini-4-argon, https://openai.com/index/gpt-6-astra/, https://epoch.ai/benchmarks/gpqa-diamond, https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash) - **3. General Human-Preference Writing, Synthesis & Hard Prompts (Arena Text)** - Best Overall: Gemini 4 Argon (1520) or Claude Opus 5.5 (1509) - Best Value: Gemini 3.8 Flash (1496, preliminary) at $0.75 / $3.75 - Best Open Weights: DeepSeek V4.1-Flash (1476, MIT) - Evidence: On Arena Text (Oct 27, 2026), four models you can use today score above 1500: Gemini 4 Argon (1520, High), Claude Opus 5.5 (1509), Muse Wonder (1507, Max) and Claude Fable 5.1 (1504). Google's gated Gemini 4 Carbon shows 1528 (preliminary). Gemini 3.8 Flash scores 1496 (preliminary) and DeepSeek V4.1-Flash 1476. (Sources: https://arena.ai/leaderboard/text) - **4. Occupational Knowledge Work, Financial Analysis & Long Documents (1M Context)** - Best Overall: Claude Opus 5.5 (1846 on GDPval-AA v2.1) - Best Value: GPT-6.1 Sol (1.05M context) - Best Open Weights: Muse Spark Open (1M context) - Evidence: Anthropic reports Claude Opus 5.5 at 1846 on GDPval-AA v2.1 (44 occupations), against 1735 for Fable 5.1. GPT-6.1 Sol has a 1.05M-token context at $2 / $10 per 1M for prompts up to 272K tokens and $4 / $15 above. Muse Spark Open has a 1M context window and open weights. (Sources: https://www.anthropic.com/claude-opus-5-5, https://developers.openai.com/api/docs/models/gpt-6.1-sol, https://developers.openai.com/api/docs/pricing, https://research.meta.ai/blog/muse-spark-open) - **5. High-Throughput Production Chat, Tool Calling & Sub-Second Loops** - Best Overall: Claude Sonnet 5.5 (low effort) - Best Value: DeepSeek V4.1-Flash (low) at $0.15 / $0.60 off-peak - Best Open Weights: DeepSeek V4.1-Flash (low) - Evidence: Artificial Analysis measures DeepSeek V4.1-Flash-1017 at 174 tok/s, Claude Sonnet 5.5 at 155, Muse Spark Open at 153 (varies by host), Gemini 3.8 Flash at 117 and Claude Opus 5.5 at 110. Lower effort means fewer reasoning tokens before the answer; Anthropic's effort guide describes lower effort as faster and cheaper for simpler tasks. (Sources: https://artificialanalysis.ai/models/deepseek-v4-1-flash-1017, https://artificialanalysis.ai/models/claude-sonnet-5-5, https://artificialanalysis.ai/models/muse-spark-open, https://platform.claude.com/docs/en/build-with-claude/effort) - **6. Open-Weights Self-Hosting, On-Prem Privacy & Custom Fine-Tuning** - Best Overall: DeepSeek V4.1-Flash or V4-Pro (MIT License) - Best Value: DeepSeek V4.1-Flash ($0.15 / $0.60 API, or self-host under MIT) - Best Open Weights: DeepSeek V4.1-Flash (MIT) · Muse Spark Open (Muse Model License) · Muse Glimmer 30B (Apache 2.0) - Evidence: DeepSeek publishes V4.1-Flash and V4-Pro weights under the MIT license, with a 1M context and a setting to turn thinking on or off. Meta's Muse Spark Open (816B total, 42B active, 1M context, 89.0% GPQA Diamond) uses the Muse Model License with limited commercial terms; Muse Glimmer 30B is Apache 2.0. (Sources: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash, https://api-docs.deepseek.com/guides/thinking_mode, https://research.meta.ai/blog/muse-spark-open, https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) ## 5. Scheduled Price Step-Ups, Promo Expirations & Deprecation Radar - **[PRICE STEP-UP (2X)] 2026-12-01 (Google DeepMind - Gemini 4 Argon)**: Introductory $2.00 / $10.00 per 1M rate ends Nov 30, 2026; the standard rate becomes $4.00 / $20.00 per 1M on Dec 1, 2026. Action: Budget twice the Gemini 4 Argon token spend from Dec 1, or compare Gemini 3.8 Flash ($0.75 / $3.75 through Dec 31) and Claude Sonnet 5.5 ($2 / $10). (https://ai.google.dev/gemini-api/docs/pricing) - **[PROMO ENDS] 2026-11-30 (OpenAI - GPT-5.6 Sol)**: GPT-5.6 Sol's promotional $4.00 / $20.00 per 1M price runs until Nov 30, 2026. Action: Compare GPT-6.1 Sol ($2 / $10 per 1M for prompts up to 272K tokens) before the promotion ends. (https://developers.openai.com/api/docs/pricing) - **[PRICE STEP-UP (2X)] 2027-01-01 (Google DeepMind - Gemini 3.8 Flash)**: Promotional $0.75 / $3.75 per 1M rate runs through Dec 31, 2026; it becomes $1.50 / $7.50 per 1M on Jan 1, 2027. Action: Finish cost-sensitive batch runs before Dec 31, 2026, and re-compare against GPT-6.1 Terra ($1 / $5) and DeepSeek V4.1-Flash ($0.15 / $0.60 off-peak) in January. (https://ai.google.dev/gemini-api/docs/pricing) - **[API ROUTING TRANSITION] When V4.1-Pro ships (DeepSeek - DeepSeek V4-Pro (0813))**: DeepSeek stated on Sep 10, 2026 that V4-Pro API service continues until DeepSeek V4.1-Pro launches (no date announced). Action: Use deepseek-flash (V4.1-Flash-1017) for coding and general tasks, or self-host the MIT weights if you need pinned V4-Pro-0813 behavior. (https://api-docs.deepseek.com/news/news260910) - **[DEPRECATION NOTICE] 2027-04-01 (OpenAI - Older GA snapshots)**: OpenAI announced its next retirement wave for Apr 1, 2027, about six months ahead. Action: Check OpenAI's deprecation table for pinned snapshots and plan migrations before Apr 1, 2027. (https://developers.openai.com/api/docs/deprecations) - **[RETIREMENT DATES] 2027-10-07 (Anthropic - Retirement dates)**: Anthropic lists a 'not sooner than' retirement date 12 months after each model's release, for example Oct 7, 2027 for Claude Haiku 5.5. Action: Pin model IDs in production and review Anthropic's deprecation table each quarter. (https://platform.claude.com/docs/en/about-claude/model-deprecations)