Repository navigation
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d55903befe
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Please review the current head |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5b25c289f3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
@codex review |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
@codex review |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
@codex review |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
@codex review Please review the current head |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Summary
This stacked PR builds the 2026-to-date ModelMap chapter on top of the reviewed 2025 corpus.
2026 backbone
Covers major releases and transitions including the Qwen3-Max-Thinking variant; Claude Opus/Sonnet 4.6; Qwen3.5 and Qwen3.6; Gemma 4; Mistral Small 4; GPT-5.5 and GPT-5.5 Pro; DeepSeek-V4 and its Pro/Flash/Vision variants; Gemini Omni Flash and Gemini 3.5–3.8 Flash-family releases; Qwen3.7-Max; the shared Claude Fable/Mythos 5 generation; Claude Sonnet 5; GPT-5.6 Sol/Terra/Luna; Claude Opus 5; Qwen3.8-Max; the shared Claude Fable/Mythos 5.1 generation; GPT-6 Astra; DeepSeek-V4.1-Flash; Claude Opus 5.5; and Claude Sonnet 5.5.
Technical evolution
Adds canonical technique nodes for Gated Delta Network, Causal Encoder-Decoder architecture, and multi-agent orchestration. Evidence-backed edges capture Qwen3.5's concrete 397B-A17B hybrid DeltaNet/attention + sparse MoE + MTP architecture; Qwen3.6-27B's hybrid thinking/non-thinking modes; Gemma 4's sparse-MoE checkpoint; Mistral Small 4 hybrid reasoning/sparse MoE; DeepSeek-V4's continued use of DSA; GPT-5.5 Pro's parallel test-time compute; the Qwen3-Max-Thinking variant's test-time scaling and tool use; and increasingly explicit agentic tool use across GPT-5.5, Gemini 3.5–3.8 Flash releases, Qwen3.7-Max, and Claude Sonnet 5.
OpenAI variant and orchestration scope
GPT-5.5 Pro is modeled as a concrete variant of GPT-5.5 because the system card describes it as the same underlying model using a parallel test-time-compute setting. GPT-5.6 is modeled as one release with Sol, Terra, and Luna variants, with separate API-release events. The
ultrasetting is represented as a product-level orchestration entity using multi-agent orchestration rather than as a model-architecture property.Google 2026 progression
Adds Gemma 4 and its E2B, E4B, 26B MoE, 31B Dense, and later 12B artifacts. The Gemini timeline now captures the July 3.6 Flash / 3.5 Flash-Lite / specialized 3.5 Flash Cyber announcement, August 3.7 Flash, and September 3.8. Gemini 3.5 Flash Cyber is modeled as a specialized fine-tuned variant of 3.5 Flash; Gemini 3.8 is a shared foundational release with Flash and Flash Cyber variants, matching Google's explicit same-foundational-intelligence framing.
Qwen and DeepSeek lifecycle detail
Qwen3-Max-Thinking is modeled as a variant of the existing Qwen3-Max release. Qwen3.6 is modeled as an umbrella generation with concrete Plus, 35B-A3B, 27B, and Max-Preview artifacts; only Max-Preview and 27B receive dated events where exact dates are established by first-party announcements.
DeepSeek-V4-Pro and DeepSeek-V4-Flash are concrete V4 variants with separate API-release events. The chronology also captures the Jul. 31 Flash post-training update, Aug. 13 Pro GA update, Aug. 21 Flash-Vision-Exp API launch, and the Sep. 10 retirement/routing transition for prior Flash variants when V4.1-Flash launched.
Safeguard and deployment semantics
Claude Fable 5 and Mythos 5 are sibling safeguard variants of one shared underlying-model release; the same structure is used for Fable/Mythos 5.1. Temporary access changes use
access_suspended/access_restoredevents attached to scopedproduct_systemdeployment entities. Fable 5 global access, Mythos 5 trusted access, and the narrower Mythos restoration for approved US organizations therefore remain separate from model identity and lineage.Chronology and provenance
Top-level model releases have explicit family lineage and publisher provenance; variant-specific techniques and availability events remain attached to concrete variants rather than being promoted to their parent release. Model announcements, weight releases, checkpoint updates, API availability, retirement, and access-policy changes remain distinct event types. GPT-6 Astra's September 3 event is represented as a limited model release, not general API availability.
Validation
CI generates a canonical 2026 year view in addition to 2023–2025 and runs the hardened schema, cross-reference, relationship-evidence, and regression checks. Current head
1d175ffis green in run #38.Stack
Base:
atlas-2025-v0.1(PR #7)Head:
atlas-2026-v0.1