New LLM Models, Sept 2026: GPT-6 Sol, Opus 5.5, GPT-6 Astra

The main September 2026 additions to the Router One catalog — GPT-6 Sol, Claude Opus 5.5, GPT-6 Astra, Grok 4.7 and DeepSeek V4.1 Flash among them — are all still listed on 2026-09-29 and callable with one API key. In the 2026-09-29 plan response, only GPT-6 Astra (Flagship models tier of Max and Ultra) and Gemini 3.8 Flash (Standard models tier of all three plans) of the twelve text models in the table below were covered by a subscription plan; the rest billed per token from the wallet. Update (2026-10-06): in the 2026-10-06 plan response, GPT-6 Sol and GPT-6.1 Sol (the latter listed 2026-09-30) are in the Premium models tier of the Pro, Max and Ultra plans; the table's plan-coverage column and the dated entries below show the plan responses of their own dates. Update (2026-10-09): as of the 2026-10-09 catalog, Claude Fable 5.1 (anthropic/claude-fable-5.1), listed again on 2026-09-30, is no longer listed; the October 2026 new-model guide and the changelog record it with the two October listings. Update (2026-10-13): as of the 2026-10-13 catalog, Gemini 3.1 Pro Preview (google/gemini-3.1-pro-preview), listed again on 2026-09-09, is no longer listed; the October guide and the changelog record it too. Update (2026-10-15): as of the 2026-10-15 catalog, Router One lists Claude Fable 5.1 (anthropic/claude-fable-5.1) again and no longer lists GPT-5.5 (openai/gpt-5.5) or Gemini 2.5 Pro (google/gemini-2.5-pro), one of the September 1–5 listings in the table below; the October guide and the changelog record these changes. Update (2026-10-19): as of the 2026-10-19 catalog, Router One no longer lists Grok 4.3 (grok-4.3) or Grok Build 0.1 (grok-build-0.1), two of the September 1–5 listings in the table below; the October guide and the changelog record it too.

September at a glance (as of 2026-09-29)

The table shows which endpoints serve each model and whether a plan covers it, per the 2026-09-29 plan response; live rates and tier boundaries are on each model page, not here. The dated updates below log every listing and delisting as it happened.

Gemini 3.1 Flash Lite also has a preview twin, google/gemini-3.1-flash-lite-preview, with the same published specs. Plan model lists change, and the pricing page shows the live ones; a model outside every plan tier bills per token from the wallet whether or not you hold a plan. The 21 Azure, AWS and Vertex channel ids listed on 2026-09-22 are covered in the channel model ids guide.

September wrap-up (as of 2026-09-30)

The re-reads on 2026-09-25, 2026-09-27 and 2026-09-29 (14:35 UTC) each returned the same 63 ids as the 2026-09-24 catalog — 56 text and 7 image — and the same plan response on all three dates; none of them showed an id listed or delisted. Four catalog changes landed after the September 24 update:

• 2026-09-26: long-context plan quota bands on 14 ids. Eight Claude ids, Gemini 3.1 Pro Preview, Gemini 3 Flash Preview and GPT-5.5 gained bands, and GPT-5.6 Sol, GPT-5.6 Terra and GPT-6 Astra a second one; the September 26 entry in the changelog lists them. Update (2026-10-13): Gemini 3.1 Pro Preview is not in the 2026-10-13 catalog. Update (2026-10-15): GPT-5.5 is not in the 2026-10-15 catalog.
• First seen 2026-09-29: capability flags on six ids. Gemini 3.8 Flash, Gemini 3.1 Flash Lite and its preview twin, Gemini 2.5 Pro, Gemini 3 Flash Preview and Claude Fable 5 now list the chat, streaming, tool-calling and vision flags; the 2026-09-25 and 2026-09-27 reads listed none for them. Grok 4.3, Grok Build 0.1 and codex-auto-review still list none. Update (2026-10-15): Gemini 2.5 Pro is not in the 2026-10-15 catalog. Update (2026-10-19): Grok 4.3 and Grok Build 0.1 are not in the 2026-10-19 catalog.
• 2026-09-29: Claude Sonnet 5.5 listed. A re-read at 18:20 UTC returned 64 ids — 57 text and 7 image: anthropic/claude-sonnet-5.5 joined (catalog release date 2026-09-29) with a 1,048,576-token context window, text and image input, the chat, streaming, tool-calling and vision flags and one rate line across the whole window. No id left and the plan response was unchanged, so no plan tier lists it and its calls bill per token to the wallet. Claude Code 2.1.284 sends claude-sonnet-5-5 for its sonnet alias; the Claude Code setup guide covers the alias pins. The Claude Sonnet 5.5 API guide covers the model id, both endpoints and what changes from Claude Sonnet 5.
• 2026-09-30: Claude Fable 5.1 listed again; GPT-6.1 Sol listed. A re-read at 06:48 UTC returned 66 ids — 59 text and 7 image — and no id left. anthropic/claude-fable-5.1, absent from the catalog since the 2026-09-11 re-read, is listed again with a 1,048,576-token context window, text and image input and the chat, streaming, tool-calling and vision flags; as a Claude-family id it is served on Messages and Chat Completions, not on Responses. Claude Code's fable alias resolves to it (per Claude Code's model-config docs, checked 2026-09-30), so the Claude Code setup guide no longer pins that alias. Update (2026-10-09): as of the 2026-10-09 catalog Router One no longer lists Claude Fable 5.1, and the setup guide pins the alias to claude-fable-5 again. Update (2026-10-15): as of the 2026-10-15 catalog Router One lists Claude Fable 5.1 again, and the setup guide no longer pins the alias. openai/gpt-6.1-sol (catalog description "GPT-6.1 Sol") is listed with a 1,050,000-token context window, text and image input and the same four flags; as a GPT-family id it is served on Chat Completions and natively on Responses. The 2026-09-30 plan response lists neither in any tier, so both bill per token to the wallet. The GPT-6.1 Sol API guide covers the model id, both endpoints, Codex CLI's new default model and what changes from GPT-6 Sol.

For the two plan-covered models in the table, a long input draws more than one plan request: GPT-6 Astra 2 above 272,000 input tokens and 4 above 512,000, Gemini 3.8 Flash 2 above 128,000, 4 above 256,000 and 8 above 512,000. The live bands sit under the plans on pricing, and the pricing facts list every band; wallet-billed calls use the posted token rates.

September 24 availability update

The 2026-09-24 catalog contains 63 ids — 56 text and 7 image. No id left. Claude Opus 5 stays listed and stays in the Premium models tier of all three plans. Claude Code 2.1.280 and later default to Claude Opus 5.5, so through Router One they send claude-opus-5-5 unless you pin a model; the Claude Code post above covers staying on plan quota. Anthropic lists breaking changes for Claude Opus 5.5 — thinking cannot be turned off, and forced tool use returns 400. Through Router One the gateway instead rewrites legacy thinking settings to adaptive and downgrades a forced tool_choice to auto, so such a request succeeds but a tool call is no longer guaranteed; check Opus 5 code against the migration guide before switching the model string. Listing dates are catalog observation dates, not vendor launch dates, and the sections below keep the earlier September snapshots as history.

September 23 availability update

The 2026-09-23 catalog contains 62 ids — 55 text and 7 image. No id left. GPT-6 Luna is not listed, and no channel twin of GPT-6 Sol is listed (azure/gpt-6-astra was then the only GPT-6 channel id). As of the 2026-10-17 catalog, Router One lists azure/gpt-6-sol and azure/gpt-6.1-sol, Azure channel ids of GPT-6 Sol and GPT-6.1 Sol; the October 2026 new-model guide records them. Listing dates are catalog observation dates, not vendor launch dates, and the sections below keep the earlier September snapshots as history.

September 22 availability update

The 2026-09-22 catalog contains 60 ids — 53 text and 7 image. A channel id keeps the key, base URL and request shape and changes only the model string; routing stays inside the channel with no fallback to the default channel, the rates and tiers are the channel's own and do not inherit the default id's discount, a plan quota that covers a default id does not cover its channel twin unless the live plan response lists the channel id, and the response model and the Logs row carry the channel id. The bare and vendor-prefixed ids are unchanged. Listing dates are catalog observation dates, not vendor launch dates, and the sections below keep the earlier September snapshots as history.

September 11 availability update

The 2026-09-11 catalog contains 39 ids — 32 text and 7 image. With DeepSeek ids listed again, the endpoint rule reads: POST /v1/chat/completions serves every chat model; POST /v1/messages serves the currently listed Claude-family and DeepSeek ids; POST /v1/responses is native for the currently listed GPT-family and DeepSeek ids. The seven image ids — gpt-image-2, the two GPT Image 2.5 tiers, the two Gemini image entries and the two Grok Imagine entries — each carry a flat per-image unit price on their model page. Listing dates are catalog observation dates, not vendor launch dates, and the sections below keep the earlier September snapshots as history.

September 9 availability update

On September 9, the catalog listed Gemini 3.1 Pro Preview with a 1,048,576-token context window, text and image input, and chat, streaming, tool-calling and vision capability flags. Its listing date does not establish a vendor launch date or promise permanent availability. The earlier September snapshot below is retained as history. Update (2026-10-13): Gemini 3.1 Pro Preview is not in the 2026-10-13 catalog.

What was listed, and what left

Between the 2026-09-01 and 2026-09-05 catalog snapshots, Router One listed eight new model ids — Claude Fable 5.1 (listed through the 2026-09-09 snapshot and absent from the 2026-09-11 catalog), GPT-6 Astra, Gemini 3.8 Flash, Gemini 3.1 Flash Lite and its preview twin, Gemini 2.5 Pro (not in the 2026-10-15 catalog), Grok 4.3 and Grok Build 0.1 (as of the 2026-10-19 catalog, Router One no longer lists either) — while twelve ids left, and every one of them that is still listed is callable from the same API key on the endpoint family its detail page lists. Update (2026-09-30): Claude Fable 5.1 is listed again; see the September wrap-up above. Update (2026-10-09): it is not in the 2026-10-09 catalog. Update (2026-10-15): it is listed again in the 2026-10-15 catalog.

This is the September sibling of the August 2026 new-model guide, written the same way: what the live catalog exposes for each id (context window, input modalities, capability flags, price tiers), which comparison pages render the spec sheets side by side, and how to send a first request. The 2026-09-07 re-read still contained 38 ids. On 2026-09-09, the catalog contained 37 ids — 32 text and 5 image — after Gemini 3.1 Pro Preview returned and GPT-5.4 plus GPT-5.6 Luna left. On 2026-09-11 it contained 39 ids — 32 text and 7 image — after Claude Fable 5.1 and GPT-5.4 mini left and DeepSeek V4.1 Flash, DeepSeek V4 Flash, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst were listed; the September 11 update above has the details. On 2026-09-19 it contained 38 ids — 31 text and 7 image — after GPT-5.3 Codex Spark left. On 2026-09-22 it contained 60 ids — 53 text and 7 image — after Grok 4.7 and 21 sales-channel ids (five azure/gpt-, nine aws/claude-, seven vertex/claude-*) were listed; the September 22 update above has the details. On 2026-09-23 it contained 62 ids — 55 text and 7 image — after GPT-6 Sol and Grok 4.7 Build Fast were listed; the September 23 update above has the details. On 2026-09-24 it contained 63 ids — 56 text and 7 image — after Claude Opus 5.5 was listed; the September 24 update above has the details. The 2026-09-25, 2026-09-27 and 2026-09-29 (14:35 UTC) re-reads returned the same 63 ids. At 18:20 UTC on 2026-09-29 the catalog contained 64 ids — 57 text and 7 image — after Claude Sonnet 5.5 was listed (in no plan tier); the September wrap-up above has the details. At 06:48 UTC on 2026-09-30 it contained 66 ids — 59 text and 7 image — after Claude Fable 5.1 was listed again and GPT-6.1 Sol was listed (both in no plan tier); the September wrap-up above has the details. These are observations on those dates; the live catalog is the source of truth for a new request. Update (2026-10-09): Claude Fable 5.1 is not in the 2026-10-09 catalog. Update (2026-10-13): Gemini 3.1 Pro Preview, which returned on 2026-09-09, is not in the 2026-10-13 catalog. Update (2026-10-15): in the 2026-10-15 catalog Claude Fable 5.1 is listed again, and GPT-5.5 and Gemini 2.5 Pro are not listed. Update (2026-10-19): Grok 4.3 and Grok Build 0.1 are not in the 2026-10-19 catalog.

The dates in this post are catalog snapshot dates, not vendor launch dates. We do not have a clean per-model listing day for any of the eight, so this post claims none.

No longer listed as of the same snapshot: DeepSeek V4 Pro and DeepSeek V4 Flash, GLM 5.1, GLM 5.2 and GLM 5.2 Fast, MiniMax M2.7, Doubao Seed 2.0 Lite, Doubao Seedream 5.0, Gemini 3 Pro Preview, Gemini 3.1 Pro Preview, o3 and o4-mini. That is the September 5 history, not the current lineup: Gemini 3.1 Pro Preview returned by September 9 (it is not in the 2026-10-13 catalog), DeepSeek V4 Flash was listed again on 2026-09-10 alongside the new DeepSeek V4.1 Flash (both in the September 11 update above), and the other ten ids were still absent on 2026-09-11. Check the current catalog before reusing any old id; the error codes page distinguishes an unknown model from a known model with no eligible route. GPT-5.6 Luna was listed on September 7 but is absent in the September 9 snapshot. This post does not speculate about why an id leaves.

Claude Fable 5.1

Update 2026-09-11: anthropic/claude-fable-5.1 is no longer in the catalog as of the 2026-09-11 snapshot, and its two comparison pages (Claude Fable 5.1 vs Claude Opus 5, Claude Fable 5.1 vs GPT-6 Astra) were retired with it. Claude Fable 5 and Claude Opus 5 remain listed. The description below is kept as history of what the September 1–9 snapshots showed. Update (2026-09-30): anthropic/claude-fable-5.1 is listed again — see the September wrap-up above; its two comparison pages were not restored. Update (2026-10-09): it is not in the 2026-10-09 catalog; this section and the table rows that name it are history. Update (2026-10-15): it is listed again as of the 2026-10-15 catalog — its model page shows the current entry; this section stays as history.

Between the 2026-09-01 and 2026-09-09 snapshots the catalog listed Claude Fable 5.1 with the same envelope as Claude Fable 5: a 1,048,576-token context window and text plus image input. The catalog entry shipped no capability flags of its own, so this post asserted none. As a Claude-family id it answered on POST /v1/messages — the path Claude Code uses — and on POST /v1/chat/completions; the API compatibility facts state the family rule, and each listed Claude model page lists its exact endpoints.

The two comparison pages that rendered its spec sheet were retired with the id. For the in-family question of what the higher Fable rate buys against Opus 5, Claude Fable 5 vs Claude Opus 5 covers the Fable id that stayed listed, Claude Fable 5 vs GPT-5.6 Sol pairs Fable 5 with GPT-5.6 Sol across vendors, and GPT-6 Astra vs Claude Opus 5 pairs GPT-6 Astra with Opus 5. The claude-*-thinking ids in the catalog are separate entries; reasoning output on them is billed at the model's posted output rate with no separate reasoning line (pricing facts).

GPT-6 Astra

GPT-6 Astra is the first GPT-6 id in the catalog. Its entry lists a 1,050,000-token context window, text and image input, and the chat, streaming, tool-calling and vision capability flags. As a GPT-family id it is served natively on POST /v1/responses — the wire format Codex CLI speaks, covered on the Codex CLI and Responses API page — and on POST /v1/chat/completions.

One billing detail belongs next to the spec sheet: Astra carries a whole-request price tier above 272,000 input tokens, the same threshold GPT-5.6 Sol carries, so its model page shows two rate lines rather than one. The tier rule is explained below. GPT-6 Astra vs GPT-5.6 Sol is the generation-over-generation comparison, and GPT-6 Astra vs Claude Opus 5 the cross-vendor one. For setup, see the GPT-6 Astra API guide.

Gemini 3.8 Flash, Gemini 3.1 Flash Lite and Gemini 2.5 Pro

These Gemini entries were observed in the same September snapshot, each with a 1,048,576-token context window and text plus image input. As of the 2026-09-29 catalog, all four ids list the chat, streaming, tool-calling and vision flags (the 2026-09-25 and 2026-09-27 reads listed none for them):

• Gemini 3.8 Flash is the newest Flash generation in the catalog and shares the Gemini 3.7 Flash context window and input modalities (the two now list the same four capability flags — the model page is authoritative), so once you have checked the flags, moving existing Flash traffic over is a model string change.
• Gemini 3.1 Flash Lite ships as two ids, google/gemini-3.1-flash-lite and google/gemini-3.1-flash-lite-preview, with identical published specs; the un-suffixed id was the one the Gemini 3.1 Flash Lite vs GPT-5.4 mini comparison used until that pair was retired on 2026-09-11, when GPT-5.4 mini left the catalog. For a high-volume pairing that is still rendered live, start from DeepSeek V4.1 Flash vs Gemini 3.8 Flash.
• Gemini 2.5 Pro is an older generation that the September 1–5 snapshots listed. The dates in this post are catalog dates, so its presence said nothing about a vendor release. As of the 2026-10-15 catalog, Router One no longer lists Gemini 2.5 Pro; choose a Gemini id the live catalog lists.

Each Gemini id answers on POST /v1/chat/completions; the model page is authoritative for the endpoint list and the live rate. Access from mainland China for the whole family is on the Gemini API in China page.

Grok 4.3 and Grok Build 0.1

The catalog described both as xAI text models — text input only, a 500,000-token context window — and listed no capability flags for either, so this post described nothing beyond that. Grok 4.3 sat alongside Grok 4.5 and Grok 4.6; Grok 4.6 vs Grok 4.5 covers the two mainline ids that already had a comparison. Grok Build 0.1 was a new name, and the catalog entry was all we could vouch for: what the "Build" line was meant for is not something the catalog stated, so this post made no claim about purpose. The Grok Build 0.1 vs GPT-5.3 Codex Spark comparison that paired it with another id on spec and rate was retired on 2026-09-19, when openai/gpt-5.3-codex-spark left the catalog.

Both Grok ids carried a whole-request tier above 200,000 input tokens, the threshold Grok 4.6 also has. As of the 2026-10-19 catalog, Router One no longer lists Grok 4.3 (grok-4.3) or Grok Build 0.1 (grok-build-0.1); this section stays as history, and the Grok API in China page lists the Grok chat ids the catalog carries.

How the whole-request price tiers work

Five text ids in this snapshot list a second rate line — GPT-6 Astra and GPT-5.6 Sol above 272,000 input tokens, Grok 4.3, Grok Build 0.1 and Grok 4.6 above 200,000 (the FAQ below lists every id with such a line as of 2026-09-29) — and the rule is the same for all of them: the total input tokens of one request select the tier, exactly at the threshold the lower tier still applies, and once a request is strictly above it the selected tier's rates apply to the whole request, output included, rather than only to the tokens past the line; monthly volume plays no part. The pricing methodology spells this out, the cost calculator applies it per request, and the prompt caching guide shows how to read cache write and cache read counts in usage so input is not counted twice.

Where the prices are

Not here. A per-token figure printed in a post is wrong within weeks, so the model catalog and each model page carry the live rate and the tier boundaries; the pricing page states the positioning — pricing starts as low as 10% of official provider list prices (up to 90% off) on select models — and the pricing facts state the billing rules. Rates differ by id and change over time, so read them from the page, not from memory, and never from a comparison that ranks models by a number that has since moved.

Which endpoint serves which id

Endpoint families are not interchangeable. POST /v1/chat/completions serves every chat model in the catalog; POST /v1/messages serves the currently listed Claude-family and DeepSeek ids; POST /v1/responses is served natively for the currently listed GPT-family and DeepSeek ids (DeepSeek returned on 2026-09-10 with deepseek-v4.1-flash and deepseek-v4-flash) and, since 2026-09-22, for the Grok chat ids. A Claude-family id sent to /v1/responses, or a chat id outside the Claude and DeepSeek families sent to /v1/messages, is rejected with HTTP 400 invalid_request_error before any model is called, and the message names the path to use. For the ids in this post that means: GPT-6 Astra and GPT-6 Sol on Responses or Chat Completions; the Grok ids likewise on Responses or Chat Completions since 2026-09-22; the Gemini ids on Chat Completions; the two DeepSeek ids on all three; Claude Opus 5.5 (listed 2026-09-24) on Messages or Chat Completions, not Responses; and Claude Fable 5.1, which left by 2026-09-11, was listed again on 2026-09-30, was not in the 2026-10-09 catalog and is listed again in the 2026-10-15 catalog, follows the same rule — Messages or Chat Completions — as every listed Claude id does. The OpenAI-compatible API page covers the request shape, and each model page lists exactly the endpoints it serves.

How to try one in five minutes

1. Create a key. Dashboard → API Keys → Create Key. Keys look like sk-.... If this is an experiment, give the key a maxSpend cap so a runaway loop stops at a number you chose (per-key cost tracking).
2. Set the base URL. https://api.router.one/v1 for OpenAI-compatible clients and SDKs; https://api.router.one (host only) for the Anthropic SDK and for Claude Code's ANTHROPIC_BASE_URL, both of which append /v1/messages themselves.
3. Send one request per family.

Chat Completions with a GPT id:

curl https://api.router.one/v1/chat/completions \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-6-astra",
    "messages": [{"role": "user", "content": "Explain what a whole-request price tier is, in two sentences."}]
  }'

Messages with a Claude id:

curl https://api.router.one/v1/messages \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "anthropic/claude-fable-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Explain what a whole-request price tier is, in two sentences."}]
  }'

4. Read the trace. Dashboard → Logs shows both calls with model, input and output tokens, cost, latency and status, so "which of the two is cheaper per solved task" becomes a filter rather than a spreadsheet (per-request observability). Retryable upstream failures are absorbed by automatic fallback across the candidate routes for the same model, and the streaming guide covers stream: true.

Top up with a card or Alipay through one hosted checkout, or with USDT/USDC on six chains (Tron, BSC, Ethereum, Polygon, Base, Arbitrum). No US credit card required. Requests reach the catalog from mainland China without a VPN.

What to A/B first

This is not a ranking — the comparison pages render live spec sheets and rates, and your own prompt set decides. For Claude Opus 5 against Claude Fable 5 and against GPT-6 Astra, Claude Fable 5 vs Claude Opus 5 and GPT-6 Astra vs Claude Opus 5; for Claude Opus 5.5 (listed 2026-09-24) beside the Opus it follows and beside GPT-6 Sol, Claude Opus 5.5 vs Claude Opus 5 and Claude Opus 5.5 vs GPT-6 Sol; to see what the GPT-6 generation changed against GPT-5.6, GPT-6 Astra vs GPT-5.6 Sol and, since 2026-09-23, GPT-6 Sol vs GPT-5.6 Sol; for the two GPT-6 tiers side by side, GPT-6 Sol vs GPT-6 Astra; for high-volume work, DeepSeek V4.1 Flash vs Gemini 3.8 Flash and DeepSeek V4.1 Flash vs Claude Haiku 4.5; for Grok beside the Claude mid tier, Grok 4.6 vs Claude Sonnet 5. Wire the winner into Claude Code (setup), Codex CLI (setup), Cursor or Zed once — they all point at the same unified LLM API gateway; Continue connects the same way but is no longer maintained (per its README; final 2.x release 2026-06-19).

FAQ

Are DeepSeek, GLM or MiniMax models still callable through Router One? DeepSeek, yes again as of 2026-09-11: deepseek-v4.1-flash was listed on 2026-09-10, with deepseek-v4-flash listed again alongside it, and both are served on /v1/chat/completions, on /v1/messages and natively on /v1/responses. GLM and MiniMax, no. DeepSeek V4 Pro and Flash, GLM 5.1, 5.2 and 5.2 Fast, and MiniMax M2.7 left the catalog on 2026-09-05 and were still absent in the 2026-09-09 snapshot; V4 Pro, GLM and MiniMax remained unlisted on 2026-09-11. The live catalog at /models is the only source of truth: ids come and go, and a post cannot promise either direction.

Which endpoint do I use for GPT-6 Astra and the Claude ids? GPT-6 Astra is a GPT-family id, so it is served natively on /v1/responses and also on /v1/chat/completions. Claude-family ids such as Claude Opus 5.5 (listed 2026-09-24), Claude Fable 5 and Claude Opus 5 answer on /v1/messages and on /v1/chat/completions, not on /v1/responses — the rule Claude Fable 5.1 follows too (it left by 2026-09-11, was listed again on 2026-09-30, was not in the 2026-10-09 catalog and is listed again in the 2026-10-15 catalog). The two DeepSeek ids listed on 2026-09-10 are served on all three endpoints. Sending a Claude-family id to /v1/responses returns HTTP 400 invalid_request_error before any model is called, with the paths to use named in the message.

Does the long-context tier apply only to the tokens above the threshold? No. The total input tokens of one request select the tier; exactly at the threshold the lower tier still applies, and strictly above it the selected tier's rates apply to the whole request, including output. Monthly usage does not select the tier. As of the 2026-09-29 catalog, GPT-6 Astra, GPT-6 Sol (listed 2026-09-23), GPT-5.6 Sol and GPT-5.6 Terra carry the line at 272,000 input tokens, as do the five azure/gpt-* channel ids; Grok 4.3, Grok Build 0.1, Grok 4.6, Grok 4.7 and Grok 4.7 Build Fast at 200,000. As of the 2026-10-19 catalog, Router One no longer lists Grok 4.3 or Grok Build 0.1.

What do the new models cost through Router One? This post deliberately prints no per-token rates because they go stale. Each model page carries the live rate and tier boundaries, and pricing starts as low as 10% of official provider list prices (up to 90% off) on select models.

Is Gemini 2.5 Pro a new September 2026 model? No. Its appearance in the September 1–5 snapshots was a catalog listing, not a vendor release; every date in this post is a catalog snapshot date. As of the 2026-10-15 catalog, Router One no longer lists Gemini 2.5 Pro (google/gemini-2.5-pro); the Gemini API in China page lists the Gemini chat ids the catalog carries.

Next steps

• Read the August 2026 guide for the previous batch — Claude Opus 5, the GPT-5.6 family and Grok 4.6.
• Pick an id on the model catalog and check its endpoints, tier boundaries and live rate; the discount qualifier is on the pricing page.
• Moving to Claude Opus 5.5: start with the API guide, then Claude Code on Opus 5.5 for the terminal agent and the migration guide for existing code.
• Point Codex CLI at a GPT id with the Codex and Responses API guide, or Claude Code at a Claude id with the Claude Code setup.