AI Model Comparison (Sep 2026) — Price, Reasoning, Coding, Speed

Colour key: 🟪 Claude 🟩 OpenAI 🟦 Grok 🟨 GLM 🟧 DeepSeek ⬛ Composer

TL;DR

• Default pick: Claude Opus 5 — 7.6/10 capability at $5/$25, within 0.9 points of the best at half the price.
• Capability ceiling: Claude Fable 5.1 (8.5/10) — $7.17/MTok, CP value 1.2, and slowest to first token.
• Value pick: DeepSeek V4.1-Flash — CP value 34.6, 29× Fable's — $0.18/MTok, cache hits at $0.006, 194 tok/s.
• Astra and Fable 5.1 tie on coding — 58.2% vs 57.9% on the official Terminal-Bench 4.0 board, inside each other's error bars.
• Give your agent docs before buying a better model — an AGENTS.md lifts Sonnet 5 from 81% to 97% on Next.js Evals, matching Fable 5.1 at 1/5 the cost.

---

Status right now — 11 Sep 2026

Model	Vendor	Released	AA Intelligence rank	Still current?
Claude Fable 5.1	Anthropic	1 Sep 2026	#1 / 200 (53)	✅ 10 days old — newest frontier model here
GPT-6 Astra	OpenAI	3 Sep 2026	#1= (53)	✅ 8 days old
Claude Opus 5	Anthropic	24 Jul 2026	#7 / 200 (51)	✅ 7 weeks; retirement no sooner than 24 Jul 2027
GLM-5.3	Z.ai	14 Aug 2026	#1 open-weights (45)	✅ 4 weeks
Grok 4.6	xAI	12 Aug 2026	(44)	✅ 4 weeks
DeepSeek V4.1-Flash	DeepSeek	10 Sep 2026	(40)	✅ 1 day old
Claude Sonnet 5	Anthropic	30 Jun 2026	#42 / 200 (38)	✅ 10 weeks
Cursor Composer 2.5	Cursor	18 May 2026	not listed	⚠️ 16 weeks, no successor

1. DeepSeek V4-Pro is being retired in 3 days. From 12:00 Beijing, 14 Sep 2026, every deepseek-v4-pro request routes to V4.1-Flash and bills at Flash prices, "until V4.1 Pro is released". If you have deepseek-v4-pro pinned in code, your model changes under you this week. V4.1-Flash scores higher anyway (40 vs 36) at ~1/4 the price.
2. GLM-5.3 is open-weights but not MIT. Its bespoke glm-5.3 licence requires MaaS operators with >$10B revenue over any 12 consecutive months to pass a Z.ai security review. DeepSeek is plain MIT — no such clause.
3. Cursor Composer 2.5 has gone 16 weeks without a refresh while five competitors shipped new flagships. Cursor 3.0 (Apr 2026) and the Sep 2026 self-hosted/builds features show the product is very much alive — the model just isn't moving.

Every model here is actively served and none is an abandonment risk. Anthropic and OpenAI are trading the frontier on a ~2-week cycle; DeepSeek and Z.ai are within 8–13 index points at a fraction of the price and closing; Composer 2.5 is the only one drifting.

---

Summary table — the factors that decide it

Four capability axes scored 0–10, averaged into a single 0–10 capability score, against real USD prices. CP value = capability ÷ $/MTok — capability points per dollar. Sorted by CP value.

Model	💲 Input	💲 Output	💲 Cache hit	💲 $/MTok	🧠 Reason	💻 Coding	🌐 General	⚡ Speed	Capability /10	💰 CP value
🟧 DeepSeek V4.1-Flash	$0.30	$1.20	$0.006	$0.18	4.5	5 (est)	6	10	6.4	🏆 34.6
🟨 GLM-5.3	$1.40	$4.40	$0.26	$0.90	6.5	7	7	4.5	6.2	6.9
⬛ Composer 2.5	$0.50	$2.50	$0.20	$0.49	1 (est)	2.5 (est)	1	9 (est)	3.4	6.9
🟦 Grok 4.6	$2.00	$6.00	$0.50	$1.35	6	3.5	6.5	4	5.0	3.7
🟪 Claude Sonnet 5	$2.00	$10.00	$0.20	$1.54	3.5	2	6.5	5	4.2	2.8
🟩 GPT-5.6 Sol	$4.00	$20.00	$0.40	$3.08	7.5	6	7.5	4.5	6.4	2.1
🟪 Claude Opus 5	$5.00	$25.00	$0.50	$3.85	9	8.5	9	4	7.6	2.0
🟪 Claude Fable 5.1	$10.00	$50.00	$0.25	$7.17	10	9.5	10	4.5	8.5	1.2
🟩 GPT-6 Astra	$10.00	$50.00	$1.00	$7.70	10	9.5	9.5	4	8.2	1.1

What "$/MTok" means: input, output and cache-hit rates are three different prices, so no single one lets you rank models. $/MTok collapses them into one dollar figure per million tokens using the average real-world mix — 7 : 2 : 1 cache hits : fresh input : output (0.7×cache + 0.2×input + 0.1×output). That is Artificial Analysis' standard ratio, and it reflects how production traffic actually looks: most tokens are re-read cached context, few are output.

Sanity check — this formula exactly reproduces the blended prices AA publishes for the five models it lists: Sonnet 5 $1.54, DeepSeek V4.1-Flash $0.18, GLM-5.3 $0.90, Opus 5 $3.85, Fable 5.1 $7.17.

All prices USD per 1M tokens, vendor list rates, 11 Sep 2026.

Scoring basis — all reproducible:

Axis	Formula	Source
🧠 Reasoning	(AA Intelligence Index − 30) ÷ 2.3	Artificial Analysis v4.3, max effort
💻 Coding	Terminal-Bench 4.0 % ÷ 6	Official tbench.ai board, 10 Sep 2026; (est) where absent
🌐 General	Judgement: HLE (no tools) + breadth of deployment + is it a general model at all	Mixed
⚡ Speed	4 + 6×(tok/s − 50) ÷ 150, capped at 10	AA measured output throughput
🎯 Capability	(reasoning + coding + general + speed) ÷ 4	Average of the four axes
💰 CP value	Capability ÷ $/MTok (7:2:1)	Derived from the above

DeepSeek's 34.6 is 29× Fable 5.1's 1.2, but it buys you 6.4 capability against Fable's 8.5. If your task fails below a capability threshold, a high CP score is worthless: you pay twice and still don't ship. Read CP value only after you know which models clear your bar.

At 70% of the mix it drives $/MTok more than the headline rates do, and it reorders the table: Fable 5.1 ($7.17) is now cheaper than GPT-6 Astra ($7.70) purely on its $0.25 cache read vs Astra's $1.00 — despite identical $10/$50 list prices. DeepSeek reads cache at $0.006, 42× cheaper than Fable 5.1 and 167× cheaper than Astra. If your workload has no cache reuse, use the no-cache column in the pricing section instead.

---

Variant detail — every model at every setting

The scorecard uses one row per model family. In practice you buy a variant: an effort level, a Fast tier, a peak/off-peak window. This is where the real decisions get made — and where the biggest surprises are.

Variant	Input	Output	Cache hit	$/MTok	Terminal-Bench 4.0	ARC-AGI-2	Notes
🟩 GPT-6 Astra max	$10.00	$50.00	$1.00	$7.70	58.2%	95.0%	Benchmark ceiling, $3,267/run
🟩 GPT-6 Astra xhigh	$10.00	$50.00	$1.00	$7.70	57.9%	93.3%	$2,351/run — 28% cheaper, same score
🟩 GPT-6 Astra high	$10.00	$50.00	$1.00	$7.70	57.9%	92.1%	$2,269/run — the value setting
🟩 GPT-6 Astra medium	$10.00	$50.00	$1.00	$7.70	54.2%	92.1%	$1,915/run
🟩 GPT-6 Astra low	$10.00	$50.00	$1.00	$7.70	50.6%	—	$1,557/run — still beats every non-Astra/Fable model
🟪 Claude Fable 5.1 max	$10.00	$50.00	$0.25	$7.17	57.9%	90.0%	$6,244/run
🟪 Claude Fable 5.1 xhigh	$10.00	$50.00	$0.25	$7.17	—	90.0%	ARC cost/task $3.12 vs max's $4.49
🟪 Claude Fable 5.1 high	$10.00	$50.00	$0.25	$7.17	—	88.8%	ARC cost/task $1.67 — 2.7× cheaper than max for −1.2 pts
🟪 Claude Opus 5 max	$5.00	$25.00	$0.50	$3.85	51.8%	90.4%	$5,969/run
🟪 Claude Opus 5 high (default)	$5.00	$25.00	$0.50	$3.85	—	88.3%	ARC cost/task $1.45 vs max's $2.06; Arena coding Elo 1533 > max's 1525
🟪 Claude Opus 5 fast mode	$10.00	$50.00	—	—	—	—	Research preview, 2.5× output tok/s, Opus 5 / 4.8 only
🟪 Claude Sonnet 5 max	$2.00	$10.00	$0.20	$1.54	12.4%	—	$9,604/run — the most expensive run on the whole board, for last place
🟪 Claude Haiku 4.5	$1.00	$5.00	$0.10	$0.77	—	—	200K ctx, no effort parameter
🟩 GPT-5.6 Sol max	$4.00	$20.00	$0.40	$3.08	37.3%	92.5%	ARC-AGI-2 beats Opus 5 and Fable 5.1
🟩 GPT-5.6 Terra max	$2.00	$12.00	$0.20	$1.74	21.5%	—	Mid tier
🟩 GPT-5.6 Luna max	$0.20	$1.20	$0.02	$0.17	17.3%	—	$347/run — cheapest run on the board
🟨 GLM-5.3 max	$1.40	$4.40	$0.26	$0.90	41.8%	—	Under Claude Code, not a native agent; $2,728/run
🟨 GLM-5.3-Flash	$0.15	$0.50	$0.03	$0.10	—	—	Arena coding Elo 1534 — above GLM-5.3-max's 1528
🟦 Grok 4.6 high <200K	$2.00	$6.00	$0.50	$1.35	20.3%	67.1%	Under xAI's own Grok Build harness
🟦 Grok 4.6 high ≥200K	$4.00	$12.00	$1.00	$2.70	—	—	Whole request reprices at the threshold — 2×
🟧 DeepSeek V4.1-Flash peak	$0.30	$1.20	$0.006	$0.18	31.2%*	—	*Self-reported; absent from the official board
🟧 DeepSeek V4.1-Flash off-peak	$0.15	$0.60	$0.003	$0.09	31.2%*	—	50% off outside 01–04 & 06–10 UTC Mon–Fri
🟧 DeepSeek V4-Pro	$1.32	$3.96	$0.044	$0.69	~27% (est)	61.3%	Reroutes to V4.1-Flash 14 Sep
⬛ Composer 2.5	$0.50	$2.50	$0.20	$0.49	~15% (est)	—	9.3 min/task; Next.js Evals 81% at $0.062/eval
⬛ Composer 2.5 Fast	$3.00	$15.00	$0.50	$2.45	—	—	6.7 min/task (30% faster) for 5× the $/MTok

Four things this table shows that the family-level scorecard hides:

1. Effort is a bigger lever than model choice on Astra. high scores 57.9% — identical to max's 58.2% inside the CI — for 31% less run cost. And Astra at low (50.6%) still beats every model here except Fable 5.1 and Opus 5 at max. Paying for max on Astra buys essentially nothing.
2. Higher effort is not monotonic. Opus 5 at high beats Opus 5 at max on Arena coding Elo (1533 vs 1525), and GLM-5.3-Flash beats GLM-5.3-max there too (1534 vs 1528). Max effort is not a free upgrade — sweep it, don't assume it.
3. Composer 2.5 Fast is a bad trade on paper. 5× the token price for a 30% latency cut. Take it only when a developer is watching the cursor blink; never for batch work.
4. Sonnet 5 at max is the worst value on the board. 12.4% on Terminal-Bench 4.0 for a $9,604 run — more than Fable 5.1 ($6,244) which scored 4.7× higher. Sonnet 5 is a fine general model; it is not an agentic-coding model, and max effort makes it expensive without making it good.

---

Pick one in 30 seconds

	🟪 Opus 5 — the default	🟪 Fable 5.1 — the ceiling	🟧 DeepSeek V4.1-Flash — the value
Advantage	51 index at $3.85/MTok · TB-4.0 51.8% · Next.js Evals 94% · cache reads $0.50	#1 index (53) · TB-4.0 57.9% · Next.js Evals 97% at $0.72/eval · cache reads $0.25	$0.18/MTok — 39× cheaper than Fable · 194 tok/s · 0.96s TTFT · MIT weights · 384K max output
Don't use it if…	❌ your evals at xhigh still fall short❌ you need >194 tok/s❌ your budget is per-token not per-outcome	❌ you have a hard token budget❌ you need first token in <5s❌ ZDR is required (400s without Anthropic authorisation)	❌ you need frontier agentic coding (31.2% TB-4.0)❌ you can't send data to a PRC-hosted API❌ you need vision-heavy enterprise SLAs

	🟩 GPT-6 Astra	🟨 GLM-5.3	🟦 Grok 4.6	⬛ Composer 2.5
Advantage	TB-4.0 60% — best measured coding · 1.05M ctx · 27k output tok/task vs Fable's 78k	Best open-weights (45) · $0.90/MTok · 2.9s TTFT · $18/mo coding plan	$1.35/MTok · GDPval-AA 1663 (beats Astra's 1555) · 500K ctx	$0.49/MTok · ~200 tok/s class · $0.07/task on AA's Coding Agent Index
Don't use it if…	❌ $7.70/MTok is too much — the priciest here❌ >272K prompts (2× input, 1.5× output)❌ you need published SWE-bench numbers	❌ you're a >$10B-revenue MaaS operator (licence review)❌ you need frontier reasoning❌ you need a US/EU-hosted first-party API	❌ agentic coding matters (TB-4.0 20.3%)❌ you cross 200K tokens (whole request reprices 2×)❌ you need 1M context	❌ you need an API — there isn't one❌ you work outside Cursor❌ you need any non-coding capability

Anthropic's cache-read pricing. Fable 5.1 reads cache at 2.5% of input ($0.25/1M) vs Opus 5's 10% ($0.50/1M). Push cache reuse above the 70% the 7:2:1 ratio assumes and the 2× list-price gap between them keeps narrowing — Anthropic estimates ~25% lower cost for typical workloads and up to ~45% for highly agentic ones. Measure your own cache-hit rate before trusting any single-number price.

---

Price — the real numbers

All figures are vendor list price, USD per 1M tokens, as published on 11 Sep 2026.

Two mixes are shown. $/MTok (7:2:1) is the realistic one used everywhere else in this doc — cache hits dominate production traffic. No-cache (3:1) is the worst case: a cold prompt every call, nothing reused. The gap between the two columns is the value of prompt caching.

Model	Input	Output	Cache hit	$/MTok (7:2:1)	No-cache (3:1)	Caching saves	Context	Max output
🟪 Claude Fable 5.1	$10.00	$50.00	$0.25	$7.17	$20.00	64%	1M	128K
🟩 GPT-6 Astra	$10.00	$50.00	$1.00	$7.70	$20.00	62%	1.05M	128K
🟪 Claude Opus 5	$5.00	$25.00	$0.50	$3.85	$10.00	62%	1M	128K
🟩 GPT-5.6 Sol	$4.00	$20.00	$0.40	$3.08	$8.00	62%	1.05M	128K
🟪 Claude Sonnet 5	$2.00	$10.00	$0.20	$1.54	$4.00	62%	1M	128K
🟦 Grok 4.6	$2.00	$6.00	$0.50	$1.35	$3.00	55%	500K	not published
🟨 GLM-5.3	$1.40	$4.40	$0.26	$0.90	$2.15	58%	1M	128K
🟪 Claude Haiku 4.5	$1.00	$5.00	$0.10	$0.77	$2.00	62%	200K	64K
🟧 DeepSeek V4-Pro (retiring 14 Sep)	$1.32	$3.96	$0.044	$0.69	$1.98	65%	1M	384K
⬛ Composer 2.5	$0.50	$2.50	$0.20	$0.49	$1.00	51%	~200K (unofficial)	not published
🟧 DeepSeek V4.1-Flash	$0.30	$1.20	$0.006	$0.18	$0.53	66%	1M	384K
🟩 GPT-5.6 Luna	$0.20	$1.20	$0.02	$0.17	$0.45	62%	1.05M	128K
🟨 GLM-5.3-Flash	$0.15	$0.50	$0.03	$0.10	$0.24	58%	1M	not published

Four pricing traps that aren't in the headline rate:

Trap	Model	Cost
Long-context surcharge	GPT-6 Astra	>272K input bills 2× input, 1.5× output
Threshold repricing	Grok 4.6	≥200K tokens reprices the entire request at $4/$12
Cache write premium	GPT-6 Astra	Cache writes $12.50/1M — 25% above input
Off-peak discount	DeepSeek	50% off outside 01:00–04:00 and 06:00–10:00 UTC Mon–Fri → $0.09/MTok

Subscription alternatives (when per-token isn't how you buy):

Plan	Price	What you get
GLM Coding Lite	$18/mo ($12.60 yearly)	10,000 credits/week, GLM-5.3 + Flash
Cursor Pro	$20/mo	Composer 2.5 + all third-party models
GLM Coding Pro	$80/mo ($56 yearly)	6× Lite quota, priority access
GLM Coding Max	$168/mo ($117.60 yearly)	14× Lite quota, dedicated peak resources
Cursor Ultra	$200/mo	Highest allowance

GLM off-peak usage costs 50% of the standard credit rate (off-peak = outside Mon–Fri 14:00–18:00 UTC+8). DeepSeek publishes no coding subscription.

---

Reasoning — Artificial Analysis Intelligence Index v4.3

The only index measuring all of these on one methodology. Max effort unless noted.

Rank	Model	Index	vs leader	Cost per index point ($/MTok ÷ index)
1=	🟪 Claude Fable 5.1	53	—	$0.135
1=	🟩 GPT-6 Astra	53	—	$0.145
3	🟪 Claude Opus 5	51	−2	$0.076
4	🟩 GPT-5.6 Sol	47	−6	$0.066
5	🟨 GLM-5.3	45	−8	$0.020
6	🟦 Grok 4.6 (high)	44	−9	$0.031
7	🟧 DeepSeek V4.1-Flash	40	−13	$0.0046
8	🟪 Claude Sonnet 5	38	−15	$0.041
9	🟧 DeepSeek V4-Pro	36	−17	$0.019
—	⬛ Composer 2.5	~32 (est)	−21	$0.015

The headline is the last column. DeepSeek V4.1-Flash buys an index point for 0.46 cents against Fable 5.1's 13.5 cents — a 29× efficiency gap. The frontier costs what it costs; the question is whether the last 13 index points change your outcome.

Reasoning depth, where vendors publish it:

Benchmark	Fable 5.1	Opus 5	Astra	Grok 4.6	DS V4.1-Flash	GLM-5.3
Humanity's Last Exam (no tools)	60.9%	56.6%	55%	44%	36.8%	not published
Humanity's Last Exam (with tools)	65.0%	63.6%	not published	not published	not published	62.5%
GPQA Diamond	not published	not published	not published	not published	90.9%	not published
SciCode	not published	not published	56%	53%	not published	not published
AA-Omniscience	not published	not published	43	29	not published	not published

Anthropic, OpenAI, xAI and Z.ai publish none of them for their current flagships. Only DeepSeek still reports GPQA Diamond (90.9%). Any 2026 comparison built on SWE-bench Verified is comparing models on a benchmark their makers stopped reporting.

---

Coding — Terminal-Bench 4.0 (official leaderboard)

The full board from tbench.ai, updated 10 Sep 2026. 330 trials per row, resolution rate, with each run's total cost. This supersedes the vendor-quoted figures — it is one harness-declared board rather than nine vendor press releases.

#	Model	Agent harness	Effort	Score (95% CI)	Run cost
1	🟩 GPT-6 Astra	Codex	max	58.2% ± 2.8	$3,267
2	🟪 Claude Fable 5.1	Claude Code	max	57.9% ± 3.8	$6,244
2	🟩 GPT-6 Astra	Codex	xhigh	57.9% ± 2.7	$2,351
2	🟩 GPT-6 Astra	Codex	high	57.9% ± 3.0	$2,269
5	🟩 GPT-6 Astra	Codex	medium	54.2% ± 2.7	$1,915
6	🟪 Claude Opus 5	Claude Code	max	51.8% ± 3.4	$5,969
7	🟩 GPT-6 Astra	Codex	low	50.6% ± 2.8	$1,557
8	🟪 Claude Fable 5	Claude Code	max	44.5% ± 3.8	$7,265
9	🟨 GLM-5.3	Claude Code	max	41.8% ± 3.2	$2,728
10	🟩 GPT-5.6 Sol	Codex	max	37.3% ± 3.8	$2,542
11	🟪 Claude Opus 4.8	Claude Code	max	23.6% ± 3.6	$6,481
12	🟩 GPT-5.6 Terra	Codex	max	21.5% ± 3.3	$1,734
13	🟦 Grok 4.6	Grok Build	high	20.3% ± 3.1	$3,592
14	Gemini 3.8 Flash	mini-SWE-agent	high	19.1% ± 3.4	$1,829
15	🟩 GPT-5.6 Luna	Codex	max	17.3% ± 2.8	$347
16	🟦 Grok 4.5	Grok Build	high	12.4% ± 2.6	$2,094
16	🟪 Claude Sonnet 5	Claude Code	max	12.4% ± 3.1	$9,604
18	Gemini 3.7 Flash	mini-SWE-agent	high	11.2% ± 2.4	$1,262

Absent from the board: DeepSeek V4.1-Flash (self-reports 31.2%), DeepSeek V4-Pro, Cursor Composer 2.5, Gemini 3.1 Pro. Their rows in the scorecard stay estimated.

The harness is half the result. GLM-5.3's 41.8% is achieved under Claude Code, not a native Z.ai agent; the Gemini rows run under mini-SWE-agent; Grok runs under xAI's own Grok Build. A model's score here is a model + scaffold score, and swapping the scaffold moves it by more than most model upgrades do.

Three results worth staring at:

Finding	Numbers
Astra's effort curve is nearly flat	high 57.9% at $2,269 vs max 58.2% at $3,267 — statistically identical for 31% less
GLM-5.3 is the open-weights story	41.8% at $2,728 beats GPT-5.6 Sol's 37.3% at $2,542, and it is 4th among model families
Sonnet 5 at max is the board's worst value	12.4% for $9,604 — the most expensive run listed, 4.7× below Fable 5.1 which cost $3,360 less

The rest of the coding evidence

Benchmark	Fable 5.1	Opus 5	Astra	Grok 4.6	GLM-5.3	DS V4.1-Flash	Composer 2.5
CursorBench 3.2	73.4%	70.0%	not pub.	69.9%	not pub.	not pub.	63.2% (v3.1)
Terminal-Bench 2.1	not pub.	89.1%	not pub.	88.4%	88.2%	90.6%	66% (v2, AA)
Terminal-Bench 3.0	not pub.	not pub.	not pub.	26%	28.3%	30.0%	not pub.
DeepSWE v1.1	not pub.	not pub.	73%*	65.9%	66.9%	74.2%	not pub.
AutomationBench	31.4%	26.9%	67%*	67%*	48.2%*	54.8%*	not pub.
SWE-bench Multilingual	not pub.	not pub.	not pub.	not pub.	not pub.	not pub.	79.8%
GDPval-AA v2 (Elo)	1853	1824	1555*	1663*	1769	not pub.	not pub.

* Different harness/version — AutomationBench and GDPval-AA numbers come from AA, xAI and Z.ai runs that are not mutually comparable. Grok 4.6's GDPval-AA 1663 beating Astra's 1555 is AA's own measurement and is real; Z.ai's 1769 for GLM-5.3 is self-reported on a different run.

DeepSeek 90.6, Opus 5 89.1, Grok 4.6 88.4, GLM-5.3 88.2 — a 2.4-point spread that inverts on TB 4.0, where Opus 5 (51.8), GLM-5.3 (41.8) and Grok 4.6 (20.3) spread over 31 points. Vendors quoting TB 2.1 in Sep 2026 are quoting a solved benchmark.

Token efficiency is the sleeper coding metric. AA measured GPT-6 Astra at 27k output tokens per task vs Fable 5.1's 78k — 2.9× fewer. At identical $50/1M output pricing, that is a 2.9× real cost difference that the price table cannot show you.

---

Real-world and independent evals

Vendor benchmarks measure what vendors choose to measure. These are run by third parties on tasks that look like actual work.

Next.js Agent Evals — the most job-like benchmark here

Run by Vercel at nextjs.org/evals, last run 10 Sep 2026. 27 evals from vercel/next-evals-oss: each is a prompt plus a real Next.js app, the agent works in an isolated Vercel Sandbox, and withheld assertions run afterwards. Scored pass@4. Crucially it publishes cost and wall-clock per eval, and it is the only current board that includes Cursor Composer 2.5.

#	Model	Agent	Avg duration	Avg cost/eval	Pass	+AGENTS.md
1	🟪 Claude Fable 5.1 (high)	Claude Code	247.6s	$0.722	97%	97%
2	🟪 Claude Opus 5	Claude Code	368.4s	$1.96	94%	97%
3	🟩 GPT-6 Astra (high)	Codex	251.9s	$0.891	90%	97%
4	Kimi K3	OpenCode	352.3s	$0.281	84%	97%
5	🟨 GLM 5.2	OpenCode	395.5s	$0.152	81%	97%
6	🟪 Claude Sonnet 5	Claude Code	263.0s	$0.386	81%	97%
7	⬛ Composer 2.5	Cursor	273.6s	$0.062	81%	94%
8	MiniMax M3	OpenCode	260.1s	$0.053	81%	90%
9	Kimi K2.7 Code	OpenCode	318.4s	$0.139	74%	90%

Superseded rows (not re-run): GPT-5.6 Sol (ultra) 77% · Claude Fable 5 (high) 77% · Grok 4.6 71% · Claude Opus 4.8 68% · Gemini 3.1 Pro 58% · Composer 2.0 58%. GLM-5.3 and both DeepSeek V4.x models have never been run — not found.

This board reorders everything. Sonnet 5 scores 81% here against 12.4% on Terminal-Bench 4.0 — and Composer 2.5 matches it at 1/6 the cost ($0.062 vs $0.386 per eval), while Fable 5.1 wins outright at $0.722, a third of Opus 5's $1.96. On ordinary framework work the cheap models are much closer to the frontier than the agentic benchmarks suggest.

Handing the agent bundled Next.js docs lifts Composer 2.5 from 81% → 94% and Sonnet 5 from 81% → 97% — matching Fable 5.1's unaided score, for the price of a context file. Six of the nine current models converge on 97% once documentation is supplied. Before paying for a better model, give the one you have better docs.

LiveBench — contamination-resistant, monthly refresh

livebench.ai, release LiveBench-2026-06-25, 52 models, with cost per task.

#	Model	Overall	Reasoning	Coding	Agentic coding	Cost/task
1	🟪 Claude Fable 5.1 (max)	83.4	91.7	86.4	66.1	$1.212
2	🟪 Claude Fable 5 (max)	83.0	89.7	86.0	62.2	$1.439
3	🟩 GPT-6 Astra (max)	82.2	92.7	80.4	57.3	$0.736
4	Muse Spark 1.3 (xhigh)	81.6	89.7	81.1	64.1	$0.219
5	🟧 DeepSeek V4.1-Flash (max)	81.1	86.7	80.0	77.3	$0.029
6	🟩 GPT-5.6 Sol (max)	81.0	91.7	83.9	56.2	$0.515
8	🟪 Claude Opus 5 (max)	80.1	91.2	81.4	65.2	$0.699
12	🟦 Grok 4.6	78.0	90.5	76.8	57.0	$0.207
17	🟧 DeepSeek V4-Pro 0813	77.4	—	—	—	—
18	Gemini 3.1 Pro (high)	77.0	—	—	44.1	—
23	🟨 GLM-5.3	76.1	—	—	—	—
24	🟪 Claude Sonnet 5 (xhigh)	76.0	—	—	—	—

DeepSeek V4.1-Flash scores 77.3 on agentic coding — the highest in that column, above Fable 5.1's 66.1 and Opus 5's 65.2, at $0.029/task versus $1.212 (42× cheaper). That directly contradicts its 31.2% self-reported Terminal-Bench 4.0 and its absence from the official board. One of those two measurements is misleading and I cannot tell you which — run your own eval before betting either way. Cursor Composer 2.5 is not on LiveBench.

ARC-AGI-2 — abstract reasoning, with cost per task

arcprize.org, semi-private eval, generated 4 Sep 2026. Human panel baseline = 100% at $17/task.

Model (effort)	Score	Cost/task
🟩 GPT-6 Astra (max)	95.0%	$1.12
🟩 GPT-6 Astra (xhigh)	93.3%	$0.83
🟩 GPT-5.6 Sol (max)	92.5%	$1.44
🟩 GPT-6 Astra (high / medium)	92.1%	$0.67 / $0.48
🟪 Claude Opus 5 (max)	90.4%	$2.06
🟪 Claude Fable 5.1 (max / xhigh)	90.0%	$4.49 / $3.12
🟪 Claude Fable 5.1 (high)	88.8%	$1.67
🟪 Claude Opus 5 (high)	88.3%	$1.45
Gemini 3.1 Pro (preview)	77.1%	$0.96
🟦 Grok 4.6 (xhigh)	67.1%	$0.76
🟧 DeepSeek V4 Flash 0731 (max)	61.4%	$0.04
🟧 DeepSeek V4-Pro 0813 (max)	61.3%	$0.60
🟨 GLM-5.2	22.8%	$0.25

OpenAI owns this benchmark. Astra at medium (92.1%, $0.48/task) beats Opus 5 at max (90.4%, $2.06) and Fable 5.1 at max (90.0%, $4.49) — 4.3× and 9.4× cheaper respectively. If your workload looks like abstract pattern reasoning, this is the clearest model-choice signal in the whole document. GLM-5.3, Sonnet 5 and Composer 2.5 are unmeasured.

GPT-6 Astra scores 99.95% under ARC's "Provider Adapter" harness and 62.71% under the Standard harness — same model, same effort. Anyone quoting "99.9% on ARC-AGI-3" is quoting the scaffold, not the model. Claude Opus 5 (high) scores 30.16% Standard; Fable 5.1, Sonnet 5, GLM-5.3, DeepSeek and Composer are all unmeasured.

Human preference — Arena Elo

LMArena has moved to arena.ai. Pairwise human votes; ±CI shown.

Board	Snapshot	Top of the board
Overall text	3 Sep 2026, 8.0M votes	Fable 5 1507, Opus 4.6-high 1505, Fable 5.1-max 1504, Opus 4.7-high 1502 … Opus 5-high 1493 (#9), Opus 5-max 1488, Gemini 3.1 Pro 1487, Sol-xhigh 1483, GLM-5.3-max 1482, GLM-5.3-Flash 1474, Sonnet 5-high 1462, Grok 4.6-high 1461, DeepSeek V4-Pro 1458
Coding	2 Sep 2026, 1.68M votes	Opus 4.7-high 1552, Opus 4.6-high 1552, Fable 5 1551, Kimi K3-max 1542 … GLM-5.3-Flash 1534, Opus 5-high 1533, Sol-xhigh 1529, GLM-5.3-max 1528, Opus 5-max 1525
WebDev Arena	8 Sep 2026, 663k votes	Astra-max 1796, Fable 5.1-max 1764, Opus 5-max 1688, Qwen3.8-max 1685, Kimi K3-max 1674 … Grok 4.6-high 1624, Sol-xhigh 1617, GLM-5.3-max 1613, DeepSeek V4.1-Flash 1620, Sonnet 5-high 1538 (#30)

Humans and benchmarks disagree sharply on the overall text board — the top four are older Anthropic models (Fable 5, Opus 4.6, Opus 4.7), with Opus 5 only 9th. Vote counts explain part of it: Fable 5.1-max has 2,906 votes against Opus 4.6-high's 72,099, so the new models' intervals are 2–3× wider. Treat the text board as lagging; WebDev Arena, with a clear Astra → Fable 5.1 → Opus 5 ordering, tracks the benchmarks far better. GPT-6 Astra and Composer 2.5 are entirely absent from the text and coding boards.

Real usage — what people actually run

Adoption is not quality, but it is the one number no vendor can spin.

OpenRouter, week to 10 Sep 2026	Tokens		Vercel AI Gateway, 9–10 Sep 2026	Team share
Hy4 preview (Tencent)	19.1T		GLM-5.3-Flash	35.8%
GPT-5.6 Luna	14.2T		Grok 4.6	11.9%
DeepSeek V4 Flash 0731	12.4T		Claude Haiku 4.5	7.7%
GLM-5.3-Flash	12.3T		Gemini 2.5 Flash	6.6%
GLM-5.3	3.06T		Claude Sonnet 5	6.3%
Claude Opus 5	1.72T (#15)		GPT-5.6 Luna	6.1%
GPT-5.6 Sol	1.71T (#16)		MiniMax M3	6.0%
Claude Sonnet 5	1.37T (#19)

Real-world usage is dominated by cheap models, by an order of magnitude. GLM-5.3-Flash and DeepSeek Flash together move ~25T tokens/week on OpenRouter while Opus 5 moves 1.72T; on Vercel's gateway a single cheap model holds 35.8% of teams. The frontier wins benchmarks; the floor wins deployments. (OpenRouter's % column is week-over-week change, not share — and it excludes first-party API traffic, where Anthropic and OpenAI are strongest, so this understates them.)

METR time horizons — stale, and worth knowing why

METR measures the task length a model completes with 50% reliability — the best single proxy for long-horizon agentic capability. Last updated 8 May 2026, so not one model in this comparison is measured. The frontier as METR last saw it: Claude Mythos Preview 17.4h, Claude Opus 4.6 12.0h, Gemini 3.1 Pro 6.4h, GPT-5.4 5.7h, Claude Opus 4.5 4.9h. METR's fitted doubling time is 128.7 days (CI 104–158). Extrapolating that to Opus 5 or Fable 5.1 would be guesswork, so this doc doesn't. METR also flags that "measurements above 16 hrs are unreliable with our current task suite".

What the official SWE-bench boards do and don't show

swebench.com's newest entry is 26 Feb 2026 — no model in this comparison appears. Its apples-to-apples subset (mini-SWE-agent 2.0.0, single attempt) tops out at Claude Opus 4.5 76.8% for $376.95 per run. Scale's SWE-bench Pro is also behind: Muse Spark 1.1 61.5%, GPT-5.4 xhigh 59.1%, Claude Opus 4.6 51.9%. The widely-quoted "Fable 5.1 leads SWE-bench Pro at 81.2%" is not on Scale's official board — it appears only on aggregator sites and is excluded here.

---

Speed

Model	Output tok/s	Time to first token	Interactive feel
🟧 DeepSeek V4.1-Flash	194.3	0.96s	Instant
⬛ Composer 2.5	~200 (est)	not published	Instant (Cursor's whole pitch)
🟪 Claude Sonnet 5	73.2	212.72s	Long wait, then fast
🟪 Claude Fable 5.1	66.3	304.16s	Longest wait
🟩 GPT-5.6 Sol	59.6	163.50s	Long wait
🟨 GLM-5.3	59.4	2.91s	Responsive
🟩 GPT-6 Astra	54.4	339.78s	Longest wait
🟦 Grok 4.6	53.3	41.39s	Noticeable pause
🟪 Claude Opus 5	51.7	107.85s	Long wait

These are max-effort measurements where TTFT includes the entire thinking phase. Models that stream their reasoning (GLM-5.3 at 2.91s, DeepSeek at 0.96s) look 100× faster than models that hide it (Claude defaults to display: "omitted", OpenAI likewise) — but the hidden-reasoning models are thinking, not stalling. Drop to low/medium effort and Claude's TTFT falls to single-digit seconds. Throughput (tok/s) is the honest cross-vendor number; TTFT is an artefact of effort setting and reasoning-display policy.

That is why the Speed score above uses throughput only.

Two speed levers the table doesn't capture:

• Claude fast mode (Opus 5 / Opus 4.8, research preview): up to 2.5× output tok/s at $10/$50 — same price as Fable 5.1, at Opus capability.
• Composer 2.5 Fast: 6.7 min/task vs 9.3 min standard (~30% faster) for 6× the list rates ($3/$15 vs $0.50/$2.50) — 5× on $/MTok.

---

Estimated numbers — and the scoreboard on how they did

Where no direct benchmark existed, numbers were bridged through models appearing on both benchmarks. The official Terminal-Bench 4.0 board has since falsified two of them, which is the most useful thing in this section: it tells you how much to trust the ones still outstanding.

Estimates that have now been checked against real data

Estimate	I said	Official board says	Error	Verdict
GLM-5.3 TB 4.0	~29%	41.8%	+12.8 pts	❌ Wrong — under-rated by 31%. The TB 3.0 → 4.0 bridge ratio (1.04) did not transfer.
Claude Sonnet 5 TB 4.0	~30%	12.4%	−17.6 pts	❌ Badly wrong — over-rated by 2.4×. My "Claude models over-perform their index on terminal tasks" correction was exactly backwards for Sonnet.
Grok 4.6 TB 4.0	17% (via AA)	20.3%	+3.3 pts	✅ Close — within the harness-variation noise.

Two of three estimates missed by more than the benchmark's own spread between adjacent models. Both scorecard rows have been corrected to the measured values, and GLM-5.3's CP value rose from 6.4 to 6.9 as a result. Do not use bridged estimates to choose a model when a real measurement exists.

Estimates still outstanding

Estimate	Bridge	Working	Confidence
Composer 2.5 TB 4.0 ≈ 15%	TB 2.0 = 69.3 (Cursor) / TB v2 = 66% (AA)	TB 2.1→3.0 collapse ratio ~0.31 across GLM, DeepSeek and Grok, extrapolated below its measured range	🚨 Very low — and now doubly suspect, since the same method under-rated GLM-5.3 by 31%. Range 8–30%. Composer scores 81% on Next.js Evals, so treat this figure as measuring frontier-agentic difficulty, not everyday competence.
DeepSeek V4-Pro TB 4.0 ≈ 27%	TB 2.1 = 87.9 vs V4.1-Flash 90.6	Below V4.1-Flash on both available axes	⚠️ Low. Moot after 14 Sep — the model routes to V4.1-Flash.
Composer 2.5 AA index ≈ 32	Not on AA's index (no public API to test)	AA Coding Agent Index 62 at launch (3rd, May 2026); the Intelligence Index weights non-coding reasoning it was never trained for	🚨 Very low. Range 25–38.
Composer 2.5 ~200 tok/s	Composer 2 measured ~200 tok/s, Composer 1 ~250	Cursor published no tok/s for 2.5	⚠️ Low. Carried forward a generation.
DeepSeek V4.1-Flash TB 4.0 = 31.2%	Not an estimate — DeepSeek's own figure	Absent from the official board	⚠️ Vendor-reported, unverified. LiveBench scores its agentic coding highest of all models (77.3), which is hard to square with 31.2%.

What is not estimated: every price, context window and release date, plus every Terminal-Bench 4.0, Next.js Evals, LiveBench, ARC-AGI and Arena figure in this document.

---

How to choose

Hard blockers — these decide it outright

If…	Then
You need frontier agentic coding, price secondary	GPT-6 Astra (TB 4.0 58.2%) or Fable 5.1 (57.9%) — a tie; pick on price
Your per-token bill is the constraint	DeepSeek V4.1-Flash — 29× better index-per-dollar
You must self-host or own the weights	DeepSeek V4.1-Flash (MIT) — GLM-5.3's licence has a revenue clause
You cannot send data to a PRC-hosted API	Rules out DeepSeek and GLM entirely
You need Zero Data Retention	Rules out Fable 5.1 (400s without Anthropic authorisation)
You work inside Cursor and want cheap fast edits	Composer 2.5 — it is genuinely good at this and nothing else
You need >500K context	Rules out Grok 4.6
You need an API	Rules out Composer 2.5 — there isn't one
You need published SWE-bench Verified numbers	Only DeepSeek still publishes comparable classics

Everyone else — weighted quiz

Start at 0. Add points:

Answer yes →	Points
Long-horizon agentic runs (>30 min autonomous)	+3 Fable 5.1 / +2 Astra
Heavily cached agent loop (same prefix every turn)	+3 Fable 5.1 (cache reads $0.25 — half of Opus 5's)
Output-heavy, low cache reuse	+2 Opus 5, +2 Astra (27k tok/task vs Fable's 78k)
High request volume, moderate difficulty	+3 DeepSeek V4.1-Flash / +2 GLM-5.3
Interactive UI where users watch tokens appear	+3 DeepSeek / +2 GLM-5.3
Non-coding knowledge work (docs, research, slides)	+2 Fable 5.1 / +2 Astra
Fixed monthly budget, not per-token	+3 GLM Coding Plan ($18/mo)
Enterprise procurement, US/EU hosting, SLAs	+2 Claude / +2 OpenAI, −3 DeepSeek/GLM

Take Claude Opus 5. It is within 0.9 capability points of the best at half the price, it is the only model here whose vendor publishes first-party Terminal-Bench 4.0, and Anthropic's own guidance is to escalate to Fable 5.1 only when your evals on Opus at higher effort measurably fall short.

---

Pros

Dimension	🟪 Claude (Fable 5.1 / Opus 5 / Sonnet 5)	🟩 OpenAI (Astra / Sol)	🟦 Grok 4.6	🟨 GLM-5.3	🟧 DeepSeek V4.1-Flash	⬛ Composer 2.5
Price	Sonnet 5 at $1.54/MTok is the cheapest credible frontier-adjacent option	—	✅ $1.35/MTok for index 44	✅ $0.90/MTok for index 45 — best open-weights value	✅ $0.18/MTok, $0.09 off-peak	✅ $0.49/MTok, $0.07/task
Reasoning	✅ Index 53 (#1/200), HLE 60.9%	✅ Index 53, AA-Omniscience 43	GDPval-AA 1663 beats Astra	✅ #1 open-weights at 45	GPQA Diamond 90.9%	—
Coding	✅ TB 4.0 57.9 / 51.8 · Next.js Evals 97% / 94% — 1st and 2nd	✅ TB 4.0 58.2%, board leader · ARC-AGI-2 95%	—	✅ TB 4.0 41.8% — 4th family, best open-weights	✅ LiveBench agentic coding 77.3 — highest of any model	✅ Next.js Evals 81% at $0.062/eval, 94% with AGENTS.md
Speed	Fast mode: 2.5× output on Opus 5	—	—	✅ 2.91s TTFT	✅ 194 tok/s, 0.96s TTFT	✅ ~200 tok/s class
Context	✅ 1M / 128K out; 300K out via batch beta	✅ 1.05M — largest	500K	✅ 1M	✅ 1M ctx, 384K max output — largest output	~200K
Caching	✅ $0.25/1M cache reads on Fable 5.1 (2.5% of input)	—	$0.50/1M	$0.26/1M	✅ $0.006/1M — 42× cheaper than Fable	$0.20/1M
Openness	—	—	—	✅ Open weights	✅ MIT, no revenue clause	—

Cons

Dimension	🟪 Claude	🟩 OpenAI	🟦 Grok 4.6	🟨 GLM-5.3	🟧 DeepSeek	⬛ Composer 2.5
Price	❌ Fable 5.1 at $7.17/MTok is second-priciest	❌ Astra $7.70/MTok — priciest here; cache writes $12.50/1M; 2× over 272K	—	—	—	❌ Fast variant is 6× standard ($3/$15)
Reasoning	Sonnet 5 at index 38 is #42/200	—	❌ −9 vs leader	❌ −8 vs leader	❌ −13 vs leader	❌ Not on any reasoning index
Coding	❌ Sonnet 5: TB 4.0 12.4% for a $9,604 run — worst value on the board	❌ Publishes no first-party benchmark numbers at all	❌ TB 4.0 20.3% — bottom third	❌ Scored under Claude Code, not a native agent	❌ Self-reports TB 4.0 31.2% but is absent from the official board	❌ Not on TB 4.0, LiveBench, Arena or ARC — 16 weeks stale
Speed	❌ 52–73 tok/s — bottom half	❌ 54 tok/s, slowest TTFT (339s)	❌ 53 tok/s	❌ 59 tok/s	—	❌ No published tok/s for 2.5
Context	—	❌ Surcharge past 272K	❌ 500K, and ≥200K reprices the whole request	—	—	❌ ~200K, unofficial; ~70–120K effective per third parties
Availability	❌ Fable 5.1 blocked under ZDR; no Priority Tier on Opus 5 / Sonnet 5 / Fable 5.1	❌ No GPT-6 mini tier	❌ Max output tokens not documented	❌ Licence review for >$10B MaaS operators	❌ V4-Pro silently reroutes 14 Sep	❌ No API. Cursor only.
Transparency	—	❌ Named 8 benchmarks at launch, published zero scores	❌ Quotes a superseded index version (61 on v4.2 vs 44 on v4.3)	❌ Active params not disclosed	—	❌ No official context window, no tok/s

---

Similarities — stop weighing these

• All eight are reasoning models with adjustable effort/thinking. None is a pure completion model.
• 1M context is table stakes — Claude, GPT, GLM and DeepSeek all ship it. Only Grok (500K) and Composer (~200K) don't.
• 128K max output on Claude, GPT and GLM. DeepSeek's 384K is the only outlier.
• All support vision except DeepSeek V4-Pro (V4.1-Flash added it).
• Prompt caching everywhere, at 0.6%–10% of input price.
• Nobody publishes SWE-bench Verified for their current flagship except DeepSeek's older cards.
• All frontier models now measure on agentic benchmarks (Terminal-Bench, GDPval, AutomationBench) rather than the 2024–25 knowledge tests.

Every model here is buying the same thing — agentic task completion — and the only real question is what you pay per completed task. Claude and OpenAI charge ~20–40× DeepSeek's rate for the top ~13 index points and ~25 points of Terminal-Bench 4.0. If your tasks fail without those points, the premium is cheap; if they don't, it's pure waste.

---

Trend — how the frontier moved in 2026

Date	Event	Index at the time
18 May 2026	Cursor Composer 2.5 — 3rd on AA Coding Agent Index at 1/10 to 1/60 the cost per task	Coding Agent Index 62
30 Jun 2026	Claude Sonnet 5	38
9 Jul 2026	GPT-5.6 family (Sol / Terra / Luna)	47 / 42 / 38
24 Jul 2026	Claude Opus 5 — ARC-AGI 3 at 3× the next-best model	51
12 Aug 2026	Grok 4.6	44
14 Aug 2026	GLM-5.3 — open weights take #1 open slot at 45	45
1 Sep 2026	Claude Fable 5.1 — cache reads cut 75% to $0.25/1M	53 🔀
3 Sep 2026	GPT-6 Astra — ties Fable 5.1 at 53, wins TB 4.0 at 60%	53 🔀
10 Sep 2026	DeepSeek V4.1-Flash — 40 at $0.18/MTok, 194 tok/s	40
14 Sep 2026	DeepSeek V4-Pro reroutes to V4.1-Flash	—

Three things the curve shows:

1. The top is a two-horse tie and it changes every fortnight. Fable 5.1 (1 Sep) and Astra (3 Sep) both landed on 53 within 48 hours. Neither lead is durable enough to architect around — build model-agnostic.
2. The open-weights gap has closed to 8 points and the price gap has not. GLM-5.3 at 45 costs 9× less than the 53s. In Jan 2026 that gap was far wider on both axes.
3. The floor is rising much faster than the premium is falling. DeepSeek V4.1-Flash buys index 40 at 0.46¢/point, while everything at index 47+ still costs 6.6–14.5¢/point. Each new cheap model lifts what you get for near-nothing; the frontier's price per point has not moved.

---

Links

	Claude	OpenAI	xAI	Z.ai	DeepSeek	Cursor
Docs	platform.claude.com	developers.openai.com	docs.x.ai	docs.z.ai	api-docs.deepseek.com	cursor.com/docs
Pricing	Pricing	Pricing	Pricing	Pricing	Pricing	Pricing
Announcement	Fable 5.1 · Opus 5	GPT-6 Astra	Grok 4.6	GLM-5.3	V4.1-Flash	Composer 2.5
Weights	—	—	—	HF (bespoke licence)	HF (MIT)	—
Subscription	claude.ai	—	—	z.ai/subscribe	—	cursor.com/pricing

Every one of them moved in the fortnight before publication.

Board	What it measures	Freshness
Terminal-Bench 4.0	Agentic coding, with harness + run cost	✅ 10 Sep 2026
Next.js Agent Evals	Real framework work, cost + duration per eval	✅ 10 Sep 2026
LiveBench	Contamination-resistant, incl. agentic coding	✅ Jun 2026 release
ARC-AGI	Abstract reasoning, cost per task	✅ 4 Sep 2026
Artificial Analysis	One index across all vendors, tok/s, TTFT	✅ continuous
Arena · coding · WebDev	Human preference Elo	✅ 2–8 Sep 2026
OpenRouter rankings · Vercel AI Gateway	Real-world adoption	✅ 10 Sep 2026
METR time horizons	Long-horizon agentic capability	❌ 8 May 2026 — stale
SWE-bench · SWE-bench Pro	Classic SWE tasks	❌ Feb 2026 — stale
Aider polyglot	Polyglot editing	❌ Nov 2025 — abandoned

Gemini 3.1 Pro lists at $2.00 / $12.00 per 1M (rising to $4/$18 above 200K), and Gemini 3.8 Flash launched 2 Sep 2026 at $0.75 / $3.75 introductory — moving to $1.50 / $7.50 on 1 Jan 2027. It sits between Sonnet 5 and GLM-5.3 on price. (Third-party pricing aggregators, not verified against Google's own pricing page — treat as indicative.)

---

Background

Skip unless you're writing an ADR.

	Vendor	First model	Business model
Claude	Anthropic	Claude 1, 2023	First-party API + Bedrock / Vertex / Foundry / AWS; claude.ai subscriptions
GPT	OpenAI	GPT-3, 2020	First-party API + Azure; ChatGPT subscriptions
Grok	xAI	Grok-1, 2023	First-party API; X Premium bundling
GLM	Z.ai (Zhipu AI)	GLM-130B, 2022	Open weights + hosted API + $18–168/mo coding plans
DeepSeek	DeepSeek	DeepSeek LLM, 2023	MIT open weights + very low-cost hosted API
Composer	Cursor (Anysphere)	Composer 1, Oct 2025	Not sold separately — bundled into Cursor's $20–200/mo plans

Architecture, where disclosed: DeepSeek V4.1-Flash is 552B backbone with 8B active on prefill / 16B on decode (763B total file size); DeepSeek V4-Pro was 1.57T total / 48B active. GLM-5.3 is 753B total, active params undisclosed. Composer 2.5 is reported as a 1.04T-total / 32B-active MoE built on a Moonshot Kimi K2.5 checkpoint (third-party reporting, not confirmed by Cursor). Anthropic, OpenAI and xAI disclose nothing.

Composer lineage: Composer 1 (Oct 2025, CursorBench 38.0) → Composer 1.5 → Composer 2 (19 Mar 2026, 61.3, and a 7× price cut to $0.50/$2.50) → Composer 2.5 (18 May 2026, 63.2 on the harder v3.1 set). The jump from 1 to 2 came from Cursor's first continued-pretraining run; 2.5 added targeted textual feedback in RL and 25× more synthetic tasks grounded in real codebases.

---

Caveats

1. Two of my three bridged estimates were wrong, and the section above says by how much. GLM-5.3 was under-rated by 12.8 points, Sonnet 5 over-rated by 17.6. Both are now corrected to measured values. The remaining estimates — all Composer 2.5 and DeepSeek V4-Pro — deserve less trust than their stated ranges imply.
2. ±3 points is the noise floor even on measured rows. The official Terminal-Bench 4.0 CIs run ±2.7 to ±3.8, so Astra's 58.2% and Fable 5.1's 57.9% are a tie, not a ranking. Anthropic and Artificial Analysis also disagree by 3–4 points running the same model on the same benchmark.
3. The harness is part of every agentic score. GLM-5.3's 41.8% is under Claude Code; Grok runs under Grok Build; Gemini under mini-SWE-agent. ARC-AGI-3 shows the extreme case — GPT-6 Astra scores 99.95% vs 62.71% on the same model at the same effort, purely on scaffold. Never compare two agentic scores without checking the harness.
4. xAI's own Grok 4.6 page claims index 61. That is AA's v4.2 index (before Terminal-Bench 4.0 replaced τ³-Banking on 7 Sep). On the current v4.3 index Grok 4.6 scores 44. Both are "correct"; they are different rulers. Every index number in this doc is v4.3.
5. OpenAI's own benchmark claims remain unverified — openai.com returned 403 to every automated fetch. Every Astra number here comes from an independent board (tbench.ai, arcprize.org, nextjs.org, livebench.ai), not from OpenAI. The viral "99.9% on ARC-AGI-3" is the Provider Adapter harness, not the model; OpenAI's claimed 96% GPQA Diamond and 74.1% DeepSWE appear only on third-party blogs and are excluded.
6. Grok 4.6's release date is disputed — x.ai and AA's article say 12 Aug 2026; AA's changelog says 21 Jul 2026. Vendor date used.
7. Composer 2.5's context window is unofficial. Cursor publishes no figure. 200K comes from third-party analysis; one source puts the effective window after truncation at 70–120K.
8. TTFT is not cross-vendor comparable — see the warning in the Speed section. It is reported for completeness, excluded from scoring.
9. CP value divides by price, so it always favours cheap models — it answers "what do I get per dollar", never "is this good enough". The capability score weights its four axes equally, which flatters fast models. Both formulas are given so you can reweight.
10. The official SWE-bench boards are too stale to use. swebench.com's newest entry is 26 Feb 2026 and Scale's SWE-bench Pro is not much newer — no model in this comparison appears on either. The circulating "Fable 5.1 leads SWE-bench Pro at 81.2%" is not on Scale's board; it is excluded.
11. DeepSeek V4.1-Flash's two coding measurements flatly contradict each other. LiveBench puts its agentic coding highest of all models (77.3); DeepSeek self-reports 31.2% on Terminal-Bench 4.0 and is absent from the official board. I have not resolved this — run your own eval before relying on either.
12. METR is 4 months stale (8 May 2026) and measures none of these models. Its 128.7-day doubling time is real; extrapolating it to Opus 5 or Fable 5.1 would not be, so this doc doesn't.
13. Arena's overall text board disagrees with every benchmark — its top four are older Anthropic models and Opus 5 sits 9th. New models there have 2,900–35,000 votes against the leaders' 72,000+, so their intervals are much wider. Treat it as lagging.

---

Sources

Claude models overview · Claude Fable 5.1 launch · Claude Opus 5 launch · OpenAI pricing · OpenAI models · GPT-6 Astra system card · xAI pricing · Grok 4.6 · GLM-5.3 blog · Z.ai pricing · GLM Coding Plan · DeepSeek pricing · DeepSeek V4.1-Flash card · Composer 2.5 · Cursor models · AA leaderboard · AA: benchmarking GPT-6 Astra · AA: Composer 2.5 Coding Agent Index · Terminal-Bench 4.0 leaderboard · Next.js Agent Evals · next-evals-oss · LiveBench · ARC Prize leaderboard · Arena text · Arena coding · WebDev Arena · Design Arena · OpenRouter rankings · Vercel AI Gateway · METR time horizons · SWE-bench · SWE-bench Pro · APIs and pages queried 11 Sep 2026.