Repository navigation
Promote flash/glm to their flagships; add deepseek-v4.1-flash - #28
Merged
Merged
Conversation
- flash / gemini-flash -> google/gemini-3.8-flash. Same $0.75/$3.75 as 3.6 and listed on Base and Solana; the alias's own rule is to follow the newest Flash generation. 3.6 stays pinnable as gemini-3.6. - glm / glm5 -> zai/glm-5.3. It was held at 5.2 only because 5.3 was Base-only; sol.blockrun.ai now lists it at the same $1.4/$4.4, which was the comment's stated condition for promoting it. - qwen-max stays on qwen3.7-max: 3.8 Max costs more ($2/$6 vs $1.475/$4.425), so moving it would raise the price without a release. - deepseek/deepseek-v4.1-flash (live on Base 2026-10-20, $0.14/$0.28, same upstream and price as deepseek-chat): priced, with explicit pins deepseek-v4.1-flash / deepseek-v4.1 / v4.1-flash. Bare deepseek stays on deepseek-chat because Solana does not list v4.1-flash yet. - vision.ts: add the six vision-tagged ids from the 2026-10-20 sync (gpt-6.1-sol, gemini-3.8-flash, deepseek-v4-flash-vision-exp, qwen3.8-flash, mimo-v2.5, mistral-large-4). Without this, an image turn on /model flash would be swapped to another model. - tests: pin the new targets, the vision flags, and that the dead nvidia DeepSeek SKUs are in no free set.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
flash/gemini-flashnow point togoogle/gemini-3.8-flash. It costs the same $0.75/$3.75 as 3.6 and is listed on both Base and Solana, and the alias's own rule is to follow the newest Flash generation. 3.6 is still reachable asgemini-3.6.glm/glm5now point tozai/glm-5.3. They were held at 5.2 only because 5.3 was Base-only. sol.blockrun.ai now lists 5.3 at the same $1.4/$4.4, which is the condition the code comment set for promoting it. 5.2 is still reachable asglm-5.2.qwen-maxis unchanged and stays on 3.7. Qwen 3.8 Max costs more ($2/$6 vs $1.475/$4.425). A bare alias that moves only when the price is the same isn't the rule here, so moving it would raise what users pay without a release.deepseek/deepseek-v4.1-flash(live on Base, $0.14/$0.28, same upstream and price asdeepseek-chatper the gateway catalog).deepseek-v4.1-flash,deepseek-v4.1andv4.1-flash.deepseekstays ondeepseek-chatbecausesol.blockrun.ai/api/v1/modelsdoes not list v4.1-flash yet. This is the same both-chains ruleglmfollowed.deepseek-v4-flash/dsv4are free aliases and stay free.router/vision.ts: adds the six vision-tagged ids from the 2026-10-20 catalog sync:gpt-6.1-sol,gemini-3.8-flash,deepseek-v4-flash-vision-exp,qwen3.8-flash,mimo-v2.5andmistral-large-4. Without this, an image turn on/model flashwould have been swapped to another model.nvidia/deepseek-v4-flash/nvidia/deepseek-v4-pro: confirmed not in any free set or routing order; both now return 400 at the gateway. A test pins this.Sources
GET https://blockrun.ai/api/v1/modelsandhttps://sol.blockrun.ai/api/v1/models, read 2026-10-20. Gatewaymodels.tsfor the deepseek-v4.1-flash parity note.Checks
npm test: 397/397 pass.npx tsc --noEmitclean, brand-numbers check up to date,node dist/index.js --helpOK.No version bump, no release.