AI Bench Index

Frontier tracking

Latest foundation model releases

A chronological catalogue combining vendor-announced releases and separately labelled third-party router listings. Dates, specifications and prices retain their source type and verification date; unavailable or conflicting fields remain explicit.

Catalogue review recorded · individual claims have separate datesStructured datasetSourceMethodology

Showing 30 of 146 matching releases.

September 2026

OpenAI

GPT-6 Astra

3 Sep 2026
Proprietary

OpenAI flagship rolling out 2026-09-03. Short-context Standard $10/$50.

Predecessor: GPT-5.6 Sol · context window unchanged

Context Window
1.05M tokens
Max Out: 128k
Token Pricing ($/M)
$10 / $50
Short-context Standard. Long context (>272k input) is 2× input / 1.5× output. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
Unchanged in the catalogue: 1,050,000 tokens
Output ceiling
Unchanged in the catalogue: 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

Google DeepMind

Gemini 3.8 Flash

2 Sep 2026
Proprietary

Current Gemini Flash flagship. Introductory $0.75/$3.75 through 2026-12-31.

Predecessor: Gemini 3.7 Flash · context window unchanged

Context Window
1.05M tokens
Max Out: 66k
Token Pricing ($/M)
$0.75 / $3.75
Introductory paid-tier rate through 2026-12-31; then $1.50/$7.50. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
Unchanged in the catalogue: 1,048,576 tokens
Output ceiling
Unchanged in the catalogue: 65,536 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

Anthropic

Claude Fable 5.1

1 Sep 2026
Proprietary

Official Anthropic list $10/$50 per 1M; cache reads $0.25/MTok. 1M context, 128k max output.

Predecessor: Claude 3.5 Sonnet · context window increased by 800,000 tokens vs predecessor

Context Window
1M tokens
Max Out: 128k
Token Pricing ($/M)
$10 / $50
Same $10/$50 as Fable 5; cache reads cut to $0.25/MTok. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
200,000 tokens → 1,000,000 tokens
Output ceiling
8,192 tokens → 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

August 2026

inclusionAI

Ling 3.0 Flash Fin

27 Aug 2026
Proprietary

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights status: no verified open-weights repository record; treated as proprietary catalogue listing. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
262k tokens
Max Out: 236k
Token Pricing ($/M)
$0.06 / $0.18
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Z.ai

GLM 5.3 Flash

26 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.31M tokens
Max Out: 131k
Token Pricing ($/M)
$0.07 / $0.25
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: zai-org/GLM-5.3-Flash). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

DeepSeek

DeepSeek V4 Flash Vision Exp

21 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 384k
Token Pricing ($/M)
$0.44 / $1.32
OpenRouter default-route rate; listing also publishes longer-context overrides. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: deepseek-ai/DeepSeek-V4-Flash-Vision-Exp). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Z.ai

GLM 5.3

18 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.31M tokens
Max Out: 944k
Token Pricing ($/M)
$1.4 / $4.4
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: zai-org/GLM-5.3). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Alibaba Cloud

Qwen3.8 27B

14 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1M tokens
Max Out: 131k
Token Pricing ($/M)
$0.42 / $3
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: Qwen/Qwen3.8-27B). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

DeepSeek

DeepSeek V4 Pro

13 Aug 2026
Open weights

V4-Pro-0813. Chart price is official off-peak cache-miss $0.66/$1.98.

Predecessor: DeepSeek V3 · context window increased by 920,576 tokens vs predecessor

Context Window
1.05M tokens
Max Out: 384k
Token Pricing ($/M)
$0.66 / $1.98
Official off-peak cache-miss. Peak hours 01:00–04:00 and 06:00–10:00 UTC are 2×. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
128,000 tokens → 1,048,576 tokens
Output ceiling
8,192 tokens → 384,000 tokens
Licence
DeepSeek Model License → Open weights (see DeepSeek V4 licence)
Weights
Unchanged in the catalogue: Available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (see DeepSeek V4 licence). Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

Google DeepMind

Gemini 3.7 Flash

13 Aug 2026
Proprietary

Previous Gemini 3 Flash. Same introductory $0.75/$3.75 through 2026-12-31.

Predecessor: Gemini 2.0 Flash · context window unchanged

Context Window
1.05M tokens
Max Out: 66k
Token Pricing ($/M)
$0.75 / $3.75
Introductory paid-tier rate through 2026-12-31; then $1.50/$7.50. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
Unchanged in the catalogue: 1,048,576 tokens
Output ceiling
8,192 tokens → 65,536 tokens
Licence
Proprietary Cloud API → Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

xAI

Grok 4.6

12 Aug 2026
Proprietary

xAI flagship. $2/$6 below 200k prompt tokens; 2× above that threshold.

Predecessor: Grok 2 · context window increased by 368,928 tokens vs predecessor

Context Window
500k tokens
Max Out: Unknown
Token Pricing ($/M)
$2 / $6
Rate for prompts under 200k tokens. Longer prompts bill $4/$12 for the full request. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
131,072 tokens → 500,000 tokens
Output ceiling
4,096 tokens → Not published
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

Alibaba Cloud

Qwen3.8 2.4T A95B

12 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 262k
Token Pricing ($/M)
$2 / $6
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: Qwen/Qwen3.8-2.4T-A95B). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

NVIDIA

Nemotron 3.5 Lightning

11 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
262k tokens
Max Out: 131k
Token Pricing ($/M)
$0.08 / $0.20
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Meta AI

Muse Glimmer 30B

9 Aug 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
131k tokens
Max Out: 118k
Token Pricing ($/M)
$0.30 / $1.1
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: meta-models/Muse-Glimmer-30B). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

July 2026

DeepSeek

DeepSeek V4 Flash

31 Jul 2026
Open weights

Official 1M context, 384k max output. Off-peak cache-miss $0.22/$0.66.

Predecessor: DeepSeek V3 · context window increased by 920,576 tokens vs predecessor

Context Window
1.05M tokens
Max Out: 384k
Token Pricing ($/M)
$0.22 / $0.66
Official off-peak cache-miss. Peak hours are 2×. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
128,000 tokens → 1,048,576 tokens
Output ceiling
8,192 tokens → 384,000 tokens
Licence
DeepSeek Model License → Open weights (see DeepSeek V4 licence)
Weights
Unchanged in the catalogue: Available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (see DeepSeek V4 licence). Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

DeepSeek

DeepSeek V4 Flash 0731

31 Jul 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.31M tokens
Max Out: 131k
Token Pricing ($/M)
$0.14 / $0.28
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: deepseek-ai/DeepSeek-V4-Flash-0731). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Thinking Machines

Inkling Small

30 Jul 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 262k
Token Pricing ($/M)
$0.45 / $1.2
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: thinkingmachines/Inkling-Small). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Anthropic

Claude Opus 5

24 Jul 2026
Proprietary

Current Opus-class API model at $5/$25. Identity confirmed against Anthropic lineup used in Fable 5.1 comparisons.

Predecessor: Claude 3 Opus · context window increased by 800,000 tokens vs predecessor

Context Window
1M tokens
Max Out: 128k
Token Pricing ($/M)
$5 / $25
Opus-class $5/$25 as stated on the Sonnet 5 announcement vs Opus 4.8/Opus 5 lineup. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
200,000 tokens → 1,000,000 tokens
Output ceiling
4,096 tokens → 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

inclusionAI

Ling 3.0 Flash

23 Jul 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
262k tokens
Max Out: 33k
Token Pricing ($/M)
$0.02 / $0.06
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: inclusionAI/Ling-3.0-flash). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Google DeepMind

Gemini 3.5 Flash Lite

21 Jul 2026
Proprietary

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights status: no verified open-weights repository record; treated as proprietary catalogue listing. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 66k
Token Pricing ($/M)
$0.30 / $2.5
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Google DeepMind

Gemini 3.6 Flash

21 Jul 2026
Proprietary

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights status: no verified open-weights repository record; treated as proprietary catalogue listing. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 66k
Token Pricing ($/M)
$0.75 / $3.75
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Thinking Machines

Inkling

17 Jul 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 472k
Token Pricing ($/M)
$1 / $4.05
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: thinkingmachines/Inkling). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Moonshot AI

Kimi K3

16 Jul 2026
Open weights

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights record references a Hugging Face repository slug; licence terms must be checked at the linked repository. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 944k
Token Pricing ($/M)
$3 / $15
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision · Self-hosted weights

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Open weights (Hugging Face: moonshotai/Kimi-K3). Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

OpenAI

GPT-5.6 Luna

9 Jul 2026
Proprietary

High-throughput GPT-5.6 tier. Permanent $0.20/$1.20 from 2026-07-30.

Predecessor: GPT-4o mini · context window increased by 922,000 tokens vs predecessor

Context Window
1.05M tokens
Max Out: 128k
Token Pricing ($/M)
$0.20 / $1.2
Permanent Luna cut from $1/$6. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
128,000 tokens → 1,050,000 tokens
Output ceiling
16,384 tokens → 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

OpenAI

GPT-5.6 Luna Pro

9 Jul 2026
Proprietary

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights status: no verified open-weights repository record; treated as proprietary catalogue listing. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 128k
Token Pricing ($/M)
$0.20 / $1.2
OpenRouter default-route rate; listing also publishes longer-context overrides. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

OpenAI

GPT-5.6 Sol

9 Jul 2026
Proprietary

GPT-5.6 flagship. Promotional $4/$20 at least through 2026-11-21.

Predecessor: OpenAI o1 · context window increased by 850,000 tokens vs predecessor

Context Window
1.05M tokens
Max Out: 128k
Token Pricing ($/M)
$4 / $20
Promotional short-context Standard at least through 2026-11-21. List was $5/$30. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
200,000 tokens → 1,050,000 tokens
Output ceiling
100,000 tokens → 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

OpenAI

GPT-5.6 Sol Pro

9 Jul 2026
Proprietary

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights status: no verified open-weights repository record; treated as proprietary catalogue listing. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
1.05M tokens
Max Out: 128k
Token Pricing ($/M)
$2 / $10
OpenRouter default-route rate; listing also publishes longer-context overrides. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

OpenAI

GPT-5.6 Terra

9 Jul 2026
Proprietary

Balanced GPT-5.6 tier. Permanent $2/$12 from 2026-07-30.

Predecessor: GPT-4o · context window increased by 922,000 tokens vs predecessor

Context Window
1.05M tokens
Max Out: 128k
Token Pricing ($/M)
$2 / $12
Permanent Terra cut from $2.50/$15. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
128,000 tokens → 1,050,000 tokens
Output ceiling
16,384 tokens → 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

June 2026

Anthropic

Claude Sonnet 5

30 Jun 2026
Proprietary

Permanent $2/$10 from 2026-08-10.

Predecessor: Claude 3.5 Sonnet · context window increased by 800,000 tokens vs predecessor

Context Window
1M tokens
Max Out: 128k
Token Pricing ($/M)
$2 / $10
Introductory $2/$10 made permanent 2026-08-10. Checked 2026-09-04.
Catalogued capabilities: Code generation · Reasoning · Tool use · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us
Context ceiling
200,000 tokens → 1,000,000 tokens
Output ceiling
8,192 tokens → 128,000 tokens
Licence
Unchanged in the catalogue: Proprietary Commercial API
Weights
Unchanged in the catalogue: Not available

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-04. Launch date and price verification are separate events.

Predecessor documentation ↗
Inspect in catalog →Vendor announcement

Google DeepMind

Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)

30 Jun 2026
Proprietary

Third-party router catalogue record; vendor lifecycle not verified, so status stays unknown rather than active or beta. Weights status: no verified open-weights repository record; treated as proprietary catalogue listing. releaseDate is the OpenRouter listing date, not a vendor announcement date. Imported 2026-09-07 from OpenRouter public listings (models API + GPQA leaderboard; model pages for leaderboard entries missing from the API). Vendor official URLs, knowledge cutoffs and lineage are not yet sourced except where a weights page is linked. Coding capability is unverified for router records and recorded as unknown, not unsupported.

Context Window
66k tokens
Max Out: 59k
Token Pricing ($/M)
$0.25 / $1.5
OpenRouter default-route rate. Router rate, may include routing markup; not a vendor list price. Checked 2026-09-07.
Catalogued capabilities: Reasoning · Vision

Capability flags describe supported features, not measured quality.

What changed—and what the specifications do not tell us

No documented predecessor pairing is available in this catalogue; a generational improvement is not inferred.

A larger context ceiling is an input capacity, not evidence of more accurate long-context reasoning. Supported modalities do not establish benchmark performance. Comparable benchmark deltas have not been established for this release brief.

Licence: Proprietary Commercial API. Model source checked: 2026-09-07. Launch date and price verification are separate events.

Inspect in catalog →Router listing

Showing 30 of 146.