LLM Leaderboard 2026
Last updated: October 25, 2026, 13:00 UTC | Update cadence: within 48 hours of any qualifying release, plus a verification sweep every Monday and Thursday, plus a full monthly re-verification | Maintained by: Axis Intelligence Research & Sarah Mitchell, AI/ML Editor — 16 active models from 8 developers · 25 cross-source observations · as of 2026-10-25 | CC BY 4.0
- Models shown
- 16
- Median blended price
- $4.00
- Self-Report Coverage
- 39.1%
- Top Terminal-Bench 4.0
- 70.6%
of 16 tracked · 8 developers
USD per 1M tokens · 3:1 input:output, list price
25 of 16×4 core scores published · baseline
Claude Sonnet 5.5 · developer-reported
| Model | Released | Access | Context | Price in / out | Blended | TB 4.0 | HLE tools | FrontierCode | GPQA | Source |
|---|---|---|---|---|---|---|---|---|---|---|
Claude Sonnet 5.5Anthropic · Closed weightsLLM-2026-010 · claude-sonnet-5-5 |
GA | 1M | $2 / $10per 1M | $4.00 | 70.6% | 64.5% | 52.1% | not reported | ScoresSpecs | |
Claude Opus 5.5Anthropic · Closed weightsLLM-2026-008 · claude-opus-5-5 |
GA | 1M | $4 / $20per 1M | $8.00 | 66.4% | 67.7% | 54.4% | not reported | ScoresSpecs | |
Kimi K3.1 ProMoonshot AI · Open weights announcedLLM-2026-017 · kimi-k3.1-proprice_pending_primary;weights_announced_not_yet_published |
GA | 1.05M | not published | not published | 65.7% | not reported | not reported | not reported | ScoresSpecs | |
GPT-6 AstraOpenAI · Closed weightsLLM-2026-005 · gpt-6-astra |
GA | 1.05M | $10 / $50per 1M | $20.00 | 57.9% | 57.2% | 53.3% | 96.0% | ScoresSpecs | |
Gemini 4 ArgonGoogle · Closed weightsLLM-2026-012 · gemini-4-argon |
GA | 1.05M | $2 / $10per 1Mlist $4 / $20 | $8.00 | 57.4% | not reported | not reported | not reported | ScoresSpecs | |
Claude Fable 5.1Anthropic · Closed weightsLLM-2026-003 · claude-fable-5-1 |
GA | 1M | $10 / $50per 1M | $20.00 | 55.8% | 65.6% | 50.3% | not reported | ScoresSpecs | |
DeepSeek-V4.1-Flash-1017DeepSeek · Open weights · MITLLM-2026-016 · deepseek-flash |
GA | 1M | $0.3 / $1.2per 1M | $0.53 | 42.3% | not reported | not reported | not reported | ScoresSpecs | |
Claude Haiku 5.5Anthropic · Closed weightsLLM-2026-014 · claude-haiku-5-5 |
GA | 1M | $0.1 / $0.5per 1M | $0.20 | 39.2% | 57.4% | 46.4% | not reported | ScoresSpecs | |
Grok 4.7SpaceXAI · Closed weightsLLM-2026-007 · grok-4.7 |
GA | 500K | $2 / $6per 1M | $3.00 | 37.6% | not reported | not reported | not reported | ScoresSpecs | |
Mistral Large 4Mistral AI · Open weights announcedLLM-2026-013 · mistral-large-4-0weights_announced_not_yet_published |
Preview | 1M | $1.36 / $4.18per 1M | $2.06 | 28.3% | not reported | not reported | not reported | ScoresSpecs | |
Gemini 3.8 FlashGoogle · Closed weightsLLM-2026-004 · gemini-3.8-flash |
GA | 1.05M | $0.75 / $3.75per 1Mlist $1.5 / $7.5 | $3.00 | 19.1% | not reported | not reported | not reported | ScoresSpecs | |
DeepSeek-V4-Pro-0813DeepSeek · Open weights · MITLLM-2026-002 · deepseek-v4-pro |
GA | 1M | $1.32 / $3.96per 1M | $1.98 | not reported | 60.0% | not reported | not reported | ScoresSpecs | |
GPT-6 LunaOpenAI · Closed weightsLLM-2026-009 · gpt-6-lunano_core_benchmark_reported |
GA | 1.05M | $0.1 / $0.5per 1M | $0.20 | not reported | not reported | not reported | not reported | ScoresSpecs | |
GPT-6.1 SolOpenAI · Closed weightsLLM-2026-011 · gpt-6.1-solno_core_benchmark_reported |
GA | 1.05M | $2 / $10per 1M | $4.00 | not reported | not reported | not reported | not reported | ScoresSpecs | |
Kimi K3Moonshot AI · Open weights · Kimi K3 LicenseLLM-2026-001 · kimi-k3 |
GA | 1.05M | $3 / $15per 1M | $6.00 | not reported | 56.0% | not reported | 93.5% | ScoresSpecs | |
Muse WonderMeta · Closed weightsLLM-2026-015 · muse-wonder-1.0no_core_benchmark_reported |
GA | 1.05M | $4 / $20per 1M | $8.00 | not reported | not reported | not reported | not reported | ScoresSpecs |
No models match those filters. Reset to see all 16.
Every score is the figure printed for that model in its own developer’s announcement or model card, at the effort setting that developer reports; Axis does not run these evaluations and does not copy scores from third-party leaderboards. not reported means the developer has not published that score — it is never estimated and never entered as zero. Developers use different harnesses and effort settings, so a column is a like-for-like name, not a like-for-like test; for that reason Axis publishes no composite score. Blended price = (3 × input + output) ÷ 4 at list price. Release events live on the AI Model Release Tracker; cost scenarios on the LLM API Cost Calculator. Full dataset: LLM Leaderboard CSV (CC BY 4.0). Cite as: Axis Intelligence Research, LLM Leaderboard 2026, 2026.
Quick answer
As of October 25, 2026, Claude Sonnet 5.5 still posts the highest developer-reported Terminal-Bench 4.0 score among 16 tracked frontier models, at 70.6%, according to Anthropic. Moonshot AI’s Kimi K3.1 Pro enters third at 65.7%. Axis Intelligence Research finds developers now publish only 39.1% of the scores on its four-benchmark core set, down from 46.4% on October 11.
What changed since October 11
- Three models added: Muse Wonder (Meta, October 12), DeepSeek-V4.1-Flash-1017 (DeepSeek, October 17) and Kimi K3.1 Pro (Moonshot AI, October 24).
- One model retired: DeepSeek-V4.1-Flash, whose
deepseek-flashAPI endpoint is now served by the October 17 snapshot. - Gemini 4 Argon reached general availability in the Gemini API on October 13 as
gemini-4-argon, with a 1,048,576-token context window. - Mistral Large 4’s launch sale ended: it now lists at $1.36 input and $4.18 output per million tokens. Weights are still not published.
- Two open items closed: Kimi K3’s release date (July 16, 2026) and the MIT license of DeepSeek-V4-Pro-0813 are now verified from primary sources.
Key findings
- Claude Sonnet 5.5 leads Terminal-Bench 4.0 at 70.6%, ahead of Claude Opus 5.5 at 66.4% and Kimi K3.1 Pro at 65.7%, according to each model’s developer.
- Developers publish 25 of the 64 core-set scores, a Self-Report Coverage of 39.1%, Axis Intelligence Research finds across 16 frontier models as of October 25, 2026 — down from 46.4% two weeks earlier.
- All three models added this cycle report one core score or none: Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017 each publish only Terminal-Bench 4.0, and Meta’s Muse Wonder publishes none of the four as figures.
- The same score is not the same number everywhere: Claude Opus 5.5’s Terminal-Bench Science 0.1 result reads 58.7% in Anthropic’s announcement and 63.3% in Google’s Gemini 4 Argon table, a 4.6-point Cross-Source Divergence.
- An open-weight model now sits within 4.9 points of the top Terminal-Bench 4.0 score: Kimi K3.1 Pro reports 65.7% against Sonnet 5.5’s 70.6%, with weights promised by November 4.
Which LLM Is the Best Right Now?
There is no single answer, and the table above is built to stop you from pretending there is one. Sort it by the column that matches your work.
On agentic terminal work, the only benchmark that most frontier labs now report about their own models, the developer-reported order is Claude Sonnet 5.5 (70.6%), Claude Opus 5.5 (66.4%), Kimi K3.1 Pro (65.7%), GPT-6 Astra (57.9%), Gemini 4 Argon (57.4%) and Claude Fable 5.1 (55.8%). Sonnet 5.5 finishing above Opus 5.5 is Anthropic’s own number, printed in its own Sonnet 5.5 announcement, and the same page cautions that Opus 5.5 remains stronger on open-ended work that needs sustained judgment. Read the score as a measurement of one harness, not a verdict on the model.
Kimi K3.1 Pro is the new name in that list. Moonshot AI’s K3.1 Pro tech blog reports 65.7% on Terminal-Bench 4.0 — and prints GPT-6 Astra at 58.2%, not the 57.9% OpenAI printed for its own model. The leaderboard keeps OpenAI’s figure in the table and logs Moonshot’s in the divergence record.
On Humanity’s Last Exam with tools, Claude Opus 5.5 reports the top figure at 67.7%, followed by Claude Fable 5.1 at 65.6% and Claude Sonnet 5.5 at 64.5%. The best open-weight figure on this column is now DeepSeek-V4-Pro-0813’s 60.0%. DeepSeek’s newer Flash snapshot reports HLE only without tools, so it leaves this column empty.
Why Axis does not publish a single overall score
Each lab runs its own evaluations, with its own harness, its own effort setting and, in several cases, its own safety fallbacks switched on. Anthropic notes that Opus 5.5 was evaluated with production safeguards that rerouted some tasks to other models; OpenAI reports “the maximum at any effort”; Google uses its highest thinking setting. Averaging those numbers into one index would merge incompatible methodologies and produce a figure that means nothing. Axis publishes the columns side by side and leaves the weighting to the reader whose workload defines it.
How Many Benchmarks Do AI Labs Actually Report?
Fewer than you would think, and fewer each cycle. Axis Intelligence Research tracks four core benchmarks — Terminal-Bench 4.0, Humanity’s Last Exam with tools, FrontierCode 1.1 Main and GPQA Diamond — and records a score only when the model’s own developer prints it.
| Core benchmark | Models whose developer reports it | Coverage | Source |
|---|---|---|---|
| Terminal-Bench 4.0 | 11 of 16 | 68.8% | Developer announcements and model cards (CSV source_url) |
| Humanity’s Last Exam, with tools | 7 of 16 | 43.8% | Same |
| FrontierCode 1.1 Main | 5 of 16 | 31.2% | Same |
| GPQA Diamond | 2 of 16 | 12.5% | Same |
| All four combined (Self-Report Coverage) | 25 of 64 | 39.1% | Axis calculation |
Self-Report Coverage fell from 46.4% on October 11 to 39.1% on October 25. The drop comes from the models themselves, not from a change in method: the three additions report one core score or none, and the retired DeepSeek-V4.1-Flash had reported three. Its successor reports one.
GPQA Diamond, for years the default science benchmark in launch posts, now appears for only two tracked models: GPT-6 Astra (96.0%) and Kimi K3 (93.5%). None of the Anthropic, Google, SpaceXAI, Mistral AI or Meta documents Axis reviewed for the tracked models prints it, and DeepSeek dropped it from its newest Flash model card. The benchmark that used to be table stakes has quietly left the table.
At the other end sit GPT-6.1 Sol, GPT-6 Luna and now Muse Wonder, with zero core-set numbers. OpenAI’s announcements present results as charts and relative claims. Meta’s Muse Wonder announcement does the same, and the figures in its model card cover Muse Wonder Full — the variant Meta does not serve through its API. A reader can see the bars; they cannot copy the numbers.
Do AI Labs Report the Same Scores for the Same Model?
Not always. When one lab prints a rival’s score in its comparison table, that figure can differ from what the rival printed about itself. Axis Intelligence Research logged every case where at least two labs’ documents give a score for the same tracked model on the same benchmark.
| Model · benchmark | Reported values (by whom) | Divergence |
|---|---|---|
| Claude Opus 5.5 · Terminal-Bench Science 0.1 | 58.7% Anthropic · 63.3% Google | 4.6 pts |
| GPT-6 Astra · Terminal-Bench Science 0.1 | 64.6% OpenAI (Astra post), Anthropic · 68.1% OpenAI (GPT-6.1 Sol post), Google | 3.5 pts |
| Claude Fable 5.1 · Terminal-Bench 4.0 | 55.8% Anthropic, OpenAI · 57.9% Google, SpaceXAI | 2.1 pts |
| Claude Fable 5.1 · HLE with tools | 65.6% Anthropic · 65.0% OpenAI | 0.6 pts |
| GPT-6 Astra · Terminal-Bench 4.0 | 57.9% OpenAI, Anthropic · 58.2% Google, Moonshot AI | 0.3 pts |
| Claude Opus 5.5 · Terminal-Bench 4.0 | 66.4% Anthropic, Google | 0 |
| Gemini 3.8 Flash · Terminal-Bench 4.0 | 19.1% Google, OpenAI | 0 |
| GPT-6 Astra · HLE with tools | 57.2% OpenAI, Anthropic | 0 |
| Claude Fable 5.1 · Terminal-Bench Science 0.1 | 52.6% Anthropic, OpenAI, Google | 0 |
Five of nine pairs diverge (55.6%); the median divergence is 0.3 points and the maximum 4.6 points. This cycle added one observation — Moonshot AI’s 58.2% for GPT-6 Astra — which matches Google’s figure and leaves every summary statistic unchanged. Google’s Gemini 4 Argon evaluation methodology states that its Terminal-Bench figures for non-Gemini models come from the benchmark authors’ public leaderboard, at each model’s best thinking level, not from the vendors’ own posts. Two labs quoting the leaderboard and two labs quoting OpenAI produce two different numbers for the same model.
The most telling row is still GPT-6 Astra on Terminal-Bench Science: OpenAI itself printed 64.6% on September 3 and 68.1% on September 29. A reader quoting the first number today is quoting a figure its own author has moved past. That is why every cell in this leaderboard carries the document it came from.
How Much Do Frontier LLMs Cost Per Million Tokens?
List prices below are the standard rates the developer publishes for its base tier, verified on October 25, 2026 unless noted. The Axis Blended Price weights input three-to-one over output, which approximates retrieval-heavy and agentic workloads where prompts dwarf replies.
| Model | Input / output, list (USD per 1M) | Axis Blended Price | Basis | Source |
|---|---|---|---|---|
| Claude Fable 5.1 | $10 / $50 | $20.00 | Standard | Anthropic |
| GPT-6 Astra | $10 / $50 | $20.00 | Prompts ≤272K tokens | OpenAI |
| Claude Opus 5.5 | $4 / $20 | $8.00 | Standard | Anthropic |
| Gemini 4 Argon | $4 / $20 | $8.00 | From December 1; $2 / $10 introductory through November 30 | |
| Muse Wonder | $4 / $20 | $8.00 | Standard; no long-context premium | Meta |
| Kimi K3 | $3 / $15 | $6.00 | Cache-miss input; last verified October 11 | Moonshot AI |
| Claude Sonnet 5.5 | $2 / $10 | $4.00 | Standard | Anthropic |
| GPT-6.1 Sol | $2 / $10 | $4.00 | Prompts ≤272K tokens | OpenAI |
| Gemini 3.8 Flash | $1.50 / $7.50 | $3.00 | From January 1, 2027; $0.75 / $3.75 introductory through December 31 | |
| Grok 4.7 | $2 / $6 | $3.00 | Standard | SpaceXAI |
| Mistral Large 4 | $1.36 / $4.18 | $2.07 | List; launch sale ended | Mistral AI |
| DeepSeek-V4-Pro-0813 | $1.32 / $3.96 | $1.98 | Peak; off-peak is half | DeepSeek |
| DeepSeek-V4.1-Flash-1017 | $0.30 / $1.20 | $0.53 | Peak; off-peak is half | DeepSeek |
| Claude Haiku 5.5 | $0.10 / $0.50 | $0.20 | Prompts ≤100K tokens | Anthropic |
| GPT-6 Luna | $0.10 / $0.50 | $0.20 | Prompts ≤272K tokens | OpenAI |
| Kimi K3.1 Pro | not published | — | No rate in a fetchable primary document as of October 25 | Moonshot AI |
The median Axis Blended Price across the 15 models with a published rate is $4.00 per million tokens, up from $3.50 on October 11. The shift comes from the mix — Muse Wonder enters at $8.00 — not from any tracked lab raising a list price. Two scheduled increases are on the calendar: Gemini 4 Argon doubles on December 1 and Gemini 3.8 Flash on January 1, 2027. Because the Axis Blended Price already uses those post-introductory rates, neither date will move it.
Kimi K3.1 Pro’s price is left empty on purpose. Moonshot’s quickstart says billing is flat with no context-length tiers, but neither its pricing documentation nor its blog prints the rate in text Axis could read. Third-party listings quote a figure; until Moonshot publishes it, the leaderboard does not.
Which Open-Weight LLMs Are on the Leaderboard?
Three tracked models have downloadable weights today: Kimi K3 (Moonshot AI, under the Kimi K3 License), DeepSeek-V4-Pro-0813 and DeepSeek-V4.1-Flash-1017 (both under the MIT License, verified from their Hugging Face repositories). Two more are announced: Mistral Large 4, whose weights Mistral promised “by the end of the month,” and Kimi K3.1 Pro, whose weights Moonshot promises by November 4. The leaderboard marks both open_announced until the files exist.
On Terminal-Bench 4.0, the newest DeepSeek Flash snapshot reports 42.3%, up from the 31.2% DeepSeek printed for the September snapshot. Every open-weight model on the board still comes from a lab outside the United States.
Which Models Have the Largest Context Windows?
Most frontier models now accept about one million tokens of input. OpenAI’s GPT-6 family lists 1,050,000 tokens; Google’s Gemini 4 Argon and Gemini 3.8 Flash, Moonshot’s Kimi K3 and K3.1 Pro, and Meta’s Muse Wonder list 1,048,576; Anthropic’s four Claude 5.x models list one million. Grok 4.7 lists 500,000.
Output limits vary far more. Gemini 3.8 Flash stops at 65,536 tokens; the Claude and GPT-6 families at 128,000; DeepSeek’s API at 384,000. Gemini 4 Argon and Kimi K3.1 Pro both allow up to 1,048,576 output tokens — Argon through a “long decode” mode, with 65,536 as its default, and K3.1 Pro with 131,072 as its default. Long input is now standard. Long output is where the models still differ.
Leaderboard entries
Reverse chronological by release date. Each entry: date · headline · summary · amount · primary source. Entries older than 30 days move to the archive block.
October 24, 2026 — Moonshot AI releases Kimi K3.1 Pro
Kimi K3.1 Pro reports 65.7% on Terminal-Bench 4.0, the third-highest developer-reported figure on the board. It lists a 1,048,576-token context window and up to 1,048,576 output tokens, with a high-speed variant, kimi-k3.1-pro-highspeed. Moonshot promises full weights by November 4. Price: not yet published in a primary document · Source: Moonshot AI
October 17, 2026 — DeepSeek releases DeepSeek-V4.1-Flash-1017
DeepSeek’s new Flash snapshot reports 42.3% on Terminal-Bench 4.0 and 42.6% on Humanity’s Last Exam without tools. It now serves the deepseek-flash API and replaces the September snapshot on this board. Weights are published on Hugging Face under the MIT License. Price: $0.30 / $1.20 per 1M at peak; half off-peak · Source: DeepSeek
October 12, 2026 — Meta releases Muse Wonder
Meta’s first frontier model is available through the Meta Model API as muse-wonder-1.0, with a 1,048,576-token context window. Its announcement presents benchmark results as charts only, so none of the four core benchmarks has a printed figure. The larger Muse Wonder Full is limited to Muse Code subscribers. Price: $4 / $20 per 1M, no long-context premium · Source: Meta
October 7, 2026 — Anthropic releases Claude Haiku 5.5
Anthropic’s smallest model posts 39.2% on Terminal-Bench 4.0, 57.4% on Humanity’s Last Exam with tools and 46.4% on FrontierCode 1.1 Main. It is Anthropic’s first Haiku model with an adjustable effort setting. The same announcement cut Claude Sonnet 5.5’s cache-read price in half. Price: $0.10 / $0.50 per 1M (prompts ≤100K tokens); $0.50 / $2.50 above · Source: Anthropic
October 6, 2026 — Mistral AI previews Mistral Large 4
Mistral AI opened a public preview of Mistral Large 4, a mixture-of-experts model its model page lists at 1.05T total and 52B active parameters, with a one-million-token context. Mistral reports 28.3% on Terminal-Bench 4.0. The launch sale has ended; weights are promised by the end of October. Price: $1.36 / $4.18 per 1M · Source: Mistral AI
September 30, 2026 — Google announces Gemini 4 Argon
Google’s new flagship reports 57.4% on Terminal-Bench 4.0. After a limited rollout to Fairwind Program defenders, it became generally available in the Gemini API on October 13 as gemini-4-argon, with a 1,048,576-token context window and up to 1,048,576 output tokens in long-decode mode. Price: $2 / $10 per 1M through November 30; $4 / $20 from December 1 · Source: Google
September 29, 2026 — OpenAI releases GPT-6.1 Sol
OpenAI’s mid-tier upgrade lists a 1,050,000-token context window and 128,000 output tokens. Its announcement reports benchmark results as charts and relative claims, so none of the four core benchmarks has a printed figure. Cached input costs $0.10 per million. Price: $2 / $10 per 1M (prompts ≤272K tokens) · Source: OpenAI
September 28, 2026 — Anthropic releases Claude Sonnet 5.5
Sonnet 5.5 reports 70.6% on Terminal-Bench 4.0 — the highest developer-reported figure on the board — plus 64.5% on Humanity’s Last Exam with tools and 52.1% on FrontierCode 1.1 Main at xhigh effort. Anthropic says Opus 5.5 remains stronger on complex open-ended work. Price: $2 / $10 per 1M · Source: Anthropic
Archive (older than 30 days)
9 archived entries
September 22, 2026 — Anthropic releases Claude Opus 5.5
Opus 5.5 reports 66.4% on Terminal-Bench 4.0 at xhigh effort, 67.7% on Humanity’s Last Exam with tools and 54.4% on FrontierCode 1.1 Main. Anthropic evaluated it with production safeguards that rerouted some tasks to older models. Input and output prices fell 20% from Opus 5. Price: $4 / $20 per 1M · Source: Anthropic
September 22, 2026 — OpenAI releases GPT-6 Luna
OpenAI’s low-cost GPT-6 model lists a 1,050,000-token context and a May 18, 2026 knowledge cutoff. Its announcement reports DeepSWE v1.1 and OSWorld 2.0 results but none of the four core benchmarks. Price: $0.10 / $0.50 per 1M (prompts ≤272K tokens) · Source: OpenAI
September 21, 2026 — SpaceXAI releases Grok 4.7
Grok 4.7 reports 37.6% on Terminal-Bench 4.0 at xHigh effort and lists a 500,000-token context window. SpaceXAI says it is built on a new, larger base model than Grok 4.6, at the same price. Price: $2 / $6 per 1M · Source: SpaceXAI
September 10, 2026 — DeepSeek releases DeepSeek-V4.1-Flash (retired October 17)
DeepSeek reported 31.2% on Terminal-Bench 4.0, 63.9% on Humanity’s Last Exam with tools and 90.9% on GPQA Diamond for this snapshot. On October 17 the deepseek-flash API moved to DeepSeek-V4.1-Flash-1017, and this row left the active table. It stays in the CSV with its scores. Price: $0.30 / $1.20 per 1M at peak · Source: DeepSeek
September 3, 2026 — OpenAI releases GPT-6 Astra
GPT-6 Astra is the only tracked model with all four core scores from its developer: 57.9% on Terminal-Bench 4.0, 57.2% on Humanity’s Last Exam with tools, 53.3% on FrontierCode 1.1 Main and 96.0% on GPQA Diamond, each the maximum at any effort. Price: $10 / $50 per 1M (prompts ≤272K tokens) · Source: OpenAI
September 2, 2026 — Google releases Gemini 3.8 Flash
Google’s Flash model reports 19.1% on Terminal-Bench 4.0 and lists a 1,048,576-token input limit with 65,536 output tokens. Its introductory price runs through December 31, 2026. Price: $0.75 / $3.75 per 1M introductory; $1.50 / $7.50 from January 1, 2027 · Source: Google
September 1, 2026 — Anthropic releases Claude Fable 5.1
Anthropic’s most capable generally available model reports 55.8% on Terminal-Bench 4.0, 65.6% on Humanity’s Last Exam with tools and 50.3% on FrontierCode 1.1 Main, per the benchmark table Anthropic published with Opus 5.5. Cybersecurity and biology requests can be routed to other models. Price: $10 / $50 per 1M · Source: Anthropic
August 13, 2026 — DeepSeek ships DeepSeek-V4-Pro-0813
The general-availability release of DeepSeek-V4-Pro reports 60.0% on Humanity’s Last Exam with tools. Weights are published on Hugging Face under the MIT License. It does not report Terminal-Bench 4.0. Price: $1.32 / $3.96 per 1M at peak; half off-peak · Source: DeepSeek
July 16, 2026 — Moonshot AI releases Kimi K3
Moonshot’s 2.8T-parameter open-weight model reports 93.5% on GPQA Diamond and 56.0% on HLE-Full with tools. Weights are published under the Kimi K3 License. Price: $3 / $15 per 1M (cache-miss input) · Source: Moonshot AI
How We Curate This List
What counts. A model enters when (1) its developer has published a primary document — announcement, model card or API documentation — that Axis fetched and read; (2) it is the developer’s current model at a flagship, mid or efficiency tier; and (3) it can be used by outside customers today, through general availability, a public preview or a limited-access program.
What does not count. Superseded tiers are retired: GPT-6 Sol left the board when OpenAI described GPT-6.1 Sol as its upgrade, and DeepSeek-V4.1-Flash left when DeepSeek moved its API to the October 17 snapshot. Image, video, audio and embedding models are out of scope. Specialist variants with separate access tiers — cyber-specific or life-sciences models — are out of scope, as are variants not served through an API, such as Muse Wonder Full. Release events themselves belong to the AI Model Release Tracker; this leaderboard holds each model’s current state, and no row duplicates a release-tracker entry.
What a score is. A benchmark cell holds the figure the model’s own developer printed for that model, with the effort setting the developer names. A rival’s figure for the same model is never used in the table; it is logged separately as a divergence observation. Scores shown only as chart bars, without a printed number, are not read off the chart. Scores produced and published by third-party aggregator leaderboards are excluded. Benchmark variants are not merged: HLE-Verified is not Humanity’s Last Exam, HLE without tools is not HLE with tools, OSWorld 2.0 is not OSWorld 2.1, Terminal-Bench 2.1 is not 4.0.
What a price is. The developer’s standard list rate for the base tier, with any promotional or introductory rate shown alongside and the threshold, end date or time-of-day condition stated. A price quoted only by a reseller or router is not used.
What Axis does not do. Axis does not run these evaluations. Every score here is self-reported. That is the honest basis of this page, and it is exactly why the Cross-Source Divergence log exists.
Methodology
Self-Report Coverage. SRC = (number of core-benchmark scores the developers publish for the active tracked models) ÷ (number of active models × 4). Core set: Terminal-Bench 4.0, Humanity’s Last Exam with tools, FrontierCode 1.1 Main, GPQA Diamond. Reading of October 25, 2026: 25 ÷ (16 × 4) = 39.1%. Baseline reading of October 11, 2026: 26 ÷ (14 × 4) = 46.4%. Retired models are excluded from the reading in which they are retired.
Cross-Source Divergence. For each model–benchmark pair reported in at least two organisations’ primary documents, CSD = highest reported value − lowest reported value, in percentage points. Reading of October 25, 2026: 9 pairs, 5 non-zero, median 0.3 points, maximum 4.6 points — unchanged from the October 11 baseline. Pairs are drawn from Terminal-Bench 4.0, HLE with tools and Terminal-Bench Science 0.1, the benchmarks on which labs most often print each other’s models.
Axis Blended Price. ABP = (3 × input price + output price) ÷ 4, in USD per million tokens, at the developer’s list price for the base tier. Introductory, sale and off-peak rates are recorded but not used in ABP, so a promotion ending does not move the metric. Models without a published list price are excluded from the median.
Limitations. Developer-reported scores use different harnesses, effort settings and safeguard configurations; a shared column name does not guarantee a shared test. Coverage measures disclosure, not capability, and moves with the model mix as well as with disclosure habits. ABP ignores cache pricing, which dominates agentic workloads. The model set is selected by tier, not by performance.
About this dataset
llm-leaderboard.csv — 42 rows: 17 model rows (16 active and 1 retired, each with specs, prices, core scores and source URLs) and 25 divergence_observation rows (one per organisation-reported score in the Cross-Source Divergence log). Every row carries source_url, retrieved_date and is_primary. Licensed CC BY 4.0. Cite as: Axis Intelligence Research, LLM Leaderboard 2026, 2026.
Review trigger: any frontier release with a primary document, any list-price change on a tracked model, any developer revising a printed score, the publication of Mistral Large 4 or Kimi K3.1 Pro weights, or Moonshot publishing a Kimi K3.1 Pro price.
Frequently asked questions
What is the best LLM in October 2026?
On developer-reported Terminal-Bench 4.0, Claude Sonnet 5.5 leads at 70.6%. On Humanity’s Last Exam with tools, Claude Opus 5.5 leads at 67.7%. On GPQA Diamond, GPT-6 Astra leads at 96.0% among the two models whose developers report it.
Which LLM is the cheapest?
Claude Haiku 5.5 and GPT-6 Luna both list at $0.10 per million input tokens and $0.50 per million output tokens, an Axis Blended Price of $0.20.
What is the best open-weight LLM?
Among models whose weights are downloadable today, DeepSeek-V4.1-Flash-1017 reports the highest Terminal-Bench 4.0 score at 42.3% and DeepSeek-V4-Pro-0813 the highest HLE-with-tools score at 60.0%. Kimi K3.1 Pro reports 65.7% on Terminal-Bench 4.0, but its weights are promised for November 4 and not yet published.
Are LLM leaderboard scores reliable?
They are what the developers published, under conditions the developers chose. Axis found that the same model’s score differs between labs’ documents in 5 of 9 cases checked, by up to 4.6 points.
How often is this leaderboard updated?
Within 48 hours of any qualifying release, plus a verification sweep every Monday and Thursday and a full monthly re-verification. Across the 53 days from September 1 to October 24, 2026, tracked labs released a qualifying model every 3.79 days on average — so a weekly cycle would leave the board stale for roughly half of each week.
Does Axis test the models itself?
No. Axis compiles developer-published figures from primary documents, computes its own disclosure, divergence and price metrics, and links every number to its source.
Change log
- October 25, 2026: Added Muse Wonder, DeepSeek-V4.1-Flash-1017 and Kimi K3.1 Pro; retired DeepSeek-V4.1-Flash; updated Gemini 4 Argon to general availability with its context window, API model ID and price end date; updated Mistral Large 4 to list price after its launch sale ended; verified Kimi K3’s release date and DeepSeek-V4-Pro-0813’s license; added one divergence observation.
- October 11, 2026: Launch with 14 models.