The clamp map is only used when reasoning.effort is sent. Drop it from
catalog rows that set omitReasoningEffort so the exported snapshot does
not advertise a dead mapping.
Paid xAI models.dev regeneration still emitted Completions-era thinking
dials for off-allowlist reasoners. Bake the no-dial policy into the
resolver/generator and refresh the exported catalog snapshot.
GLM-5.3 introduces three key API changes from GLM-5.2:
- Uniform wire-exact low/high/max reasoning_effort ladder on every host
(replacing GLM-5.2's host-specific dialects)
- Thinking can no longer be disabled (thinking.type must always be "enabled")
- Default effort is max
Changes:
- Add isGlm53ReasoningEffortModelId classifier (>=5.3, base/air/turbo, non-vision)
- getModelDefinedEfforts: GLM-5.3 returns LOW_HIGH_MAX uniformly
- impliesMandatoryReasoning: GLM-5.3 floors thinking-off to lowest effort
- deriveThinking/fillThinkingWireDefaults: defaultLevel=max for GLM-5.3
- generated-policies: pin glm-5.3 to 1M context (zai + zhipu-coding-plan)
- descriptors: zai defaultModel -> glm-5.3
- generate-models: curated seed (glm-5.3 is live but not in /models discovery)
- models.json: bundled glm-5.3 entry
- Tests: catalog thinking-metadata + AI wire-mapping (5 new tests)
Applied curated Alibaba Token Plan seeds after generic models.dev fallback so bundled capabilities cannot be overwritten by incomplete upstream metadata.
- Claude ids now alias to google-vertex suffixed entries (claude-opus-4-6@default etc.) so Antigravity follows Google's price if it diverges from Anthropic's list price; plain-id anthropic lookup remains as dangling-alias fallback.
- Regenerated models.json via gen:models.
- Added `applyAntigravityPricingFallback` to back-fill unpriced `google-antigravity` models using first-party list prices from `google` and `anthropic`.
- Mapped Antigravity model IDs to preview-id aliases (e.g., `gemini-3.1-pro` to `gemini-3.1-pro-preview`) for correct pricing resolution.
- Added test coverage in `packages/catalog/test/generated-policies.test.ts` verifying fallback pricing and preservation of billable costs.
- Rewrote fetchCodexDiscoveryModels to resolve every stored openai-codex
OAuth account via getOAuthAccesses and reuse openaiCodexModelManagerOptions'
tested union/fail-closed path, so a single narrow account can no longer
authoritatively wipe sibling-account models from the bundle.
- Restored openai-codex gpt-5.4, gpt-5.6-sol, and gpt-5.3-codex-spark bundle
entries (and gpt-5.5's contextPromotionTarget) dropped by the previous
single-account regen.
- Updated the live Codex image tool-result tests off the retired
gpt-5.2-codex id to gpt-5.5.
Fixes#6265
Ollama Cloud's deepseek-v4-pro and deepseek-v4-flash deployments reject any
output budget above 65536 with HTTP 400, despite advertising a 1M context /
384K output (ollama/ollama#16890). Ollama's /api/show never reports this cap,
so the catalog left the base models at the full context window and the dated
tag deepseek-v4-flash:0731 at a stale 8192 fallback. Pin these ids (base plus
tag variants) to min(contextWindow, 65536) at both runtime discovery and
generation; other cloud models keep their discovered limits.
Fixes#7266
- Implement the ai& provider registry entry with API-key authentication and login support.
- Add model descriptors, static model seeding, and openai-compatible model discovery for the ai& provider.
- Update the model catalog with ai& provider models, pricing, and updated provider model names.
- Add unit tests for the ai& provider environment resolution, metadata, and dynamic model mapping.
- Migrated model catalog fetching and documentation references from the models.dev API to the stencil.so well-known models endpoint.
- Added support for zstd decompression and session-based caching with ETag conditional requests and stale fallback handling.
- Updated test suites, mock URLs, and constants across catalog and coding-agent packages to target stencil.so.
- Added GMI_CLOUD_STATIC_MODELS seed wired into gen:models so a fresh
install resolves deepseek-ai/DeepSeek-V4-Flash synchronously at boot,
before async /v1/models discovery fires; live discovery stays
authoritative and replaces the seed.
- Regenerated models.json with the seeded gmi-cloud slice.
- Dropped the GeneratedProvider cast now that gmi-cloud is bundled.
- Moved gmi-cloud to the end of the /login provider order.
- Moved changelog entries from released 17.0.3 sections to Unreleased.
- Added a regression test asserting the seed covers the descriptor's
defaultModel.
kimi-code discovery mapModel and the bundled catalog hardcoded
maxTokens=32000 for every model, truncating k3/k3-256k output at ~4x
below their real 131072 ceiling and kimi-for-coding[-highspeed] below
32768. Add kimiCodeMaxTokens() (k3* -> 131072, kimi-for-coding* ->
32768, else fallback), wire it into the discovery mapper and the
generator cap policy, and correct the bundled kimi-code entries.
Fixes#6711
Changes:
- Add Amazon Bedrock catalog entries for Claude Opus 5
- Exclude unsupported jp. inference profiles during generation
- Add Bedrock prompt cache compatibility info, AWS model card sources, and regression tests
Reason / background:
- Incorporate the new model info from models.dev and align with the official AWS spec
Scope of impact:
- Model catalog generation, Bedrock compatibility info, catalog tests
fetchProviderModelsFromCatalog returning succeeded=true for an empty
discovery made every dynamicModelsAuthoritative provider drop its
models.dev and previous-snapshot rows after a flaky empty-but-200
response. Restored the fetched-models requirement for other providers;
only alibaba-token-plan treats an empty success as authoritative so the
subscribed-edition allowlist is not widened by the curated seed. Also
dropped the stray trailing newline in models.json that the generator
does not emit.
- Removed the logic that forced a minimum context window of 372k for GPT-5.6 SKUs when upstream actively reported lower values.
- Updated Codex discovery to treat 372k as a fallback only when upstream omits the context window.
Fixes#6371
Codex discovery under-reports the gpt-5.6 sol/terra/luna window: some
accounts omit `context_window`, others actively return 272000. The #5707
`?? fallback` only fired on absence, so an actively-reported 272000 passed
through and `preferDiscoveryLimit` overwrote the bundled 372K pin at
runtime, dropping compaction from 279000 to 204000 tokens.
Treat GPT_5_6_CONTEXT_WINDOW as a floor for these SKUs via Math.max so
neither omission nor active under-report regresses the real capacity;
other models keep honoring the reported value. Corrected the stale
"omits" comments in codex.ts and generated-policies.ts.
Fixes#6259
main already bundles devin/swe-1-7 (discovered live, with image input);
the text-only seed wins the earlier-sources dedup in generate-models and
downgrades the bundled entry to text-only. Carry-over from the previous
snapshot already preserves the model across keyless regens.
- Repointed the Umans models.dev descriptor to published PAYG rates.
- Backfilled authoritative discovery rows and the Flash technical alias.
- Added runtime, descriptor, and bundled catalog regression coverage.
Fixes#5733
Codex discovery falls back to DEFAULT_CONTEXT_WINDOW (272000) when upstream
omits context_window, which overwrote the previously-bundled 372000 hard
capacity for openai-codex gpt-5.6 luna/sol/terra on regen. OpenAI's Codex
model registry declares context_window = max_context_window = 372000, and a
direct Responses request with 350,317 input tokens completes, proving 272K is
not the route's hard cap.
Pinned these SKUs to 372000 in applyOpenAICatalogPolicy and refreshed the
bundled entries.
Fixes#5705
- Enabled OpenAI pro reasoning mode by integrating reasoning aliases and parameter injection.
- Expanded the model catalog with GPT-5.6 Luna, Sol, Terra, and Meta Muse Spark 1.1.
- Updated model type definitions and provider request transformers to support reasoning configurations.
- Refined model generation scripts to include new pro-reasoning aliases for OpenAI providers.
Devin exposes 6 GLM-5.2 wire UIDs. Without a collapse family, all 6
appeared as separate catalog entries and the user could inadvertently
select a quota-gated variant (glm-5-2-max, glm-5-2-none) that silently
fails with 'weekly usage quota exhausted' even though the base glm-5-2
is free.
Live verification via streamDevin confirmed:
- glm-5-2 (base, 200K): FREE — works with quota exhausted
- glm-5-2-none (200K): quota-gated — fails with 'usage quota exhausted'
- glm-5-2-max (200K): quota-gated — fails with 'usage quota exhausted'
- swe-1-6, swe-1-7, kimi-k2-7: FREE — all work with quota exhausted
Changes:
- Add GLM-5.2 collapse family routing all efforts (high/xhigh) to the
free glm-5-2 wire UID — never to the quota-gated variants
- Add GLM-5.2 1M collapse family for paid variants (glm-5-2-1m,
glm-5-2-none-1m, glm-5-2-max-1m)
- Add swe-1-7 static fallback seed (missing from catalog, free on Devin)
- Wire seed in generate-models.ts with authoritative-discovery guard
- 3 unit tests for GLM-5.2 collapse routing
- Updated generated OpenCode Go DeepSeek V4 catalog policy to use max_tokens instead of max_completion_tokens.
- Added regression coverage for deepseek-v4-flash:xhigh tool requests carrying max_tokens and reasoning_effort:max.
Fixes#4647
- Implement Baseten provider support with authentication and dynamic model discovery.
- Register Baseten in the model catalog and provider priority order.
- Expand model definitions with new DeepSeek, Kimi, NVIDIA, and Claude variants.
- Update model configurations, cost data, and provider-specific metadata.
- Removed the canonical model variant indexing, selection, and tracking logic from the model registry and resolver.
- Eliminated the `canonical` sub-command, tab view, search tokens, and equivalence configuration structures from the CLI and model selector components.
- Refined model identification, lookup, and provider fallback resolution to bind exclusively to standard, raw model IDs.
- Relocated the equivalence utility script within the catalog package to support script-only policy generation.
- Raised `GEMINI_HEADER_RUNAWAY_THRESHOLD` from 10 to 24 to avoid false-positive interrupts on legitimate, complex reasoning blocks.
- Added a regression test verifying that 10 distinct, progressing headers do not trip the detector while 24 headers still trigger it.
- Introduced a model canonicalization helper to strip redundant model compatibility fields that match defaults.
- Regenerated the models catalog JSON to eliminate over eight hundred lines of redundant compatibility specifications.
- Updated the variant collapse logic to rebuild models using the projected compatibility configurations.
- Added a missing type annotation to a test environment variable to resolve a compilation warning.
can1357