RefMOC: Frontier Model Releases

MOC: Frontier Model Releases

MOC: Frontier Model Releases

This map covers major frontier model releases: when each shipped, where it is logged here, and how the frontier has moved on benchmarks that stay comparable across releases. The pace in 2026 is the story. Anthropic shipped four flagship-tier releases between June and September, and OpenAI went from GPT-5.6 Sol to GPT-6 Astra in eight weeks.

Line charts of Terminal-Bench 4.0 and Terminal-Bench-Science 0.1 scores by release date, Jun–Sep 2026: Claude rises from Fable 5 to Opus 5.5; OpenAI from GPT-5.6 Sol to GPT-6 Astra

Benchmark progress (Jun–Sep 2026)

Most benchmark rows change between announcements. GDPval-AA moved from v2 to v2.1, which changed the Elo scale. OSWorld 2.0 changed its task set, and AutomationBench's leaderboard re-scored older models. Only Terminal-Bench 4.0 and Terminal-Bench-Science 0.1 keep the same numbers across the Fable 5.1 and Opus 5.5 tables: Opus 5 reproduces at 52.3% and 29.0% in both. The chart uses only those two.

ModelReleasedTerminal-Bench 4.0Terminal-Bench-Science 0.1
Claude Fable 52026-06-0942.0%24.7%
GPT-5.6 Sol2026-07-0937.3%22.4%
Claude Opus 52026-07-2452.3%29.0%
Claude Fable 5.12026-09-0155.8%52.6%
GPT-6 Astra2026-09-0357.9%64.6%
Claude Opus 5.52026-09-2266.4%58.7%

Scores: Fable 5 from the Fable 5.1 announcement;1 all others from the Opus 5.5 announcement,2 where the shared rows match.

Claude Mythos 5.1, which has restricted access, scores 60.9% on Terminal-Bench 4.0. The science benchmark shows the steepest change: Claude went from 29% to 53% in one release (Opus 5 → Fable 5.1), and Astra reached 65%. Anthropic now says benchmark margins at this level are "a less reliable guide to real-world differences", so read the chart for direction rather than decimals. Scores come from vendors; OpenAI figures are as reported by OpenAI.

Timeline

2026

DateModelLabNotesSource
2026-02-05Claude Opus 4.6Anthropic1M context in beta3
2026-02-17Claude Sonnet 4.6Anthropic3
2026-03gpt-5-4OpenAINative computer use, 1M context4
2026-04gemma-4Google DeepMindOpen models built from Gemini 3 research5
2026-04-08meta-muse-spark-mslMetaMuse Spark, from Meta Superintelligence Labs6
2026-04Claude Mythos PreviewAnthropicRestricted; announced with anthropic-glasswing7
2026-04-16Claude Opus 4.7Anthropic3
2026-04-23gpt-5-5OpenAIAgentic coding, computer use8
2026-04deepseek-v4-previewDeepSeekOpen-weights MoE, 1M context, very low API prices9
2026-05-28Claude Opus 4.8AnthropicNow the fallback model for Opus 5.5's cyber safeguards3 2
2026-06-09Claude Fable 5 + Mythos 5AnthropicAccess revoked Jun 12 after a US government letter; see fable-5-oneshot-sakura10 11
2026-06-30Claude Sonnet 5Anthropic$2 / $10 per M tokens3
2026-07-09GPT-5.6 (Luna / Terra / Sol)OpenAILimited preview from Jun 2612
2026-07-24Claude Opus 5Anthropic1M context, adaptive thinking; see trq212-claude-5-context-engineering3
2026-09-01Claude Fable 5.1 + Mythos 5.1AnthropicMost capable general release at the time1
2026-09-03gpt-6-astraOpenAISaturates ARC-AGI-3; first "Critical" cyber rating13 14
2026-09-22claude-opus-5-5AnthropicFable 5.1-level for about 40% less than Opus 5; Sonnet/Haiku 5.5 "in the coming weeks"2

2024–2025 (Claude, for context)

All dates from the public Claude release timeline.3

DateModel
2024-03-04Claude 3 (Haiku, Sonnet, Opus)
2024-06-21Claude 3.5 Sonnet
2024-10-22Claude 3.5 Haiku; upgraded 3.5 Sonnet (computer use)
2025-02-24Claude 3.7 Sonnet (hybrid reasoning)
2025-05-22Claude Opus 4 + Sonnet 4
2025-08-05Claude Opus 4.1
2025-09-29Claude Sonnet 4.5
2025-10-15Claude Haiku 4.5
2025-11-24Claude Opus 4.5

Sources

Where a row links to a reference entry here, that entry holds the fuller write-up and links.

Footnotes

  1. Introducing Claude Fable 5.1 and Claude Mythos 5.1, Anthropic. ↩ ↩2

  2. Introducing Claude Opus 5.5, Anthropic, 2026-09-22. ↩ ↩2 ↩3

  3. anthropic-claude-timeline: a community-maintained timeline of Claude model releases. It is a secondary source; check against Anthropic's announcements where it matters. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7

  4. Introducing GPT-5.4, OpenAI. ↩

  5. Gemma 4, Google DeepMind. ↩

  6. Introducing Muse Spark, Meta. ↩

  7. Project Glasswing, Anthropic. ↩

  8. Introducing GPT-5.5, OpenAI. ↩

  9. DeepSeek V4 Preview release, DeepSeek API docs. ↩

  10. Introducing Claude Fable 5 and Claude Mythos 5, Claude Platform docs. ↩

  11. Claude Mythos, Wikipedia: covers the US government letter and the access revocation. ↩

  12. GPT-5.6, Wikipedia. ↩

  13. GPT-6 Astra: A new generation of intelligence, OpenAI. ↩

  14. GPT-6 Astra, Wikipedia: gives the Sep 3 release to approved users and general availability the next day. ↩

Edit this page

Links to this page:

  1. 2026-09-23 Claude Opus 5.5
  2. Claude Opus 5.5
  3. GPT-6 Astra: A New Generation of Intelligence